CLIN. CHEM. 38/6, 864-872 (1992)

864 CLINICALCHEMISTRY, Vol.38, No. 6, 1992

BiologicalVariability of Lipoproteinsand Apolipoproteinsin Patients Referred to aLipid ClinicStephanie D. Kafonek,”2 Carol A. Derby,4 and Paul S. Bachorik”3’5

We determined the physiological variability of total cho-lesterol, high- (HDL) and low-density lipoprotein (LDL)cholesterol, triglycerides, and apolipoproteins A-I and B infasting blood samples from patients referred to the JohnsHopkins Lipid Referral Clinic. Samples were taken oneach of three occasions during baseline evaluation visitsbefore the patients were treated. The median physiolog-ical coefficients of variation (CV) were as follows: totalcholesterol, 5.0%; triglycerides, 17.8%; HDL cholesterol,7.1%; LDL cholesterol, calculated from the previousthreemeasurements, 7.8%; and apolipoproteins A-I and B,7.1% and 6.4%, respectively. There were no significantdifferences in CVI, between children (�18 years) andadults (>18 years) for any of the measurements. On thebasis of our findings, single measurements in serialsamplestakenonthreeoccasions suffice to establish thepatients’ usual values with the following precisions (± 1CV): total cholesterol, ± 4%; triglycerides, ± 11%; HDLcholesterol, ± 8%; LDL cholesterol,± 6%; and apolipo-proteins A-I and B, ± 7%.

AddItIonal Keyphrases: cholesterol . screening baselinevalues intra-indMdual variation

The relationship between increased concentrations ofplasma total cholesterol and coronary heart disease(CHD) is well established, and the extent of risk attrib-utable to hypercholesterolemia can be defined furtherby determining the distribution of cholesterol amongthe lipoproteins (1). Increased plasma concentrations ofboth low-density-lipoprotein cholesterol (LDL-C) and itsmajor apolipoprotein, apoB, are correlated with an in-creased risk for CHD (2); and low concentrations ofhigh-density-lipoprotein cholesterol (HDL-C) and itsmajor apolipoprotein, apoA-I, are inversely correlatedwith risk (3, 4). Guidelines for the evaluation andtreatment of increased concentrations of total choles-terol and LDL-C have been established by the NationalCholesterol Education Program (5), but similar recom-mendations for HDL-C are not yet available. Severalstudies have indicated that apoB and apoA-I are strongpredictors of CHD (2) and, as measures of these apolipo-

Lipid Research-Atherosclerosis Division,’ Departments of Pedi-atrics, 2 Medicine, Laboratory Medicine, The Johns HopkinsUniversity School of Medicine,Baltimore, MD 21205.

4Pawtucket Heart Health Program, The Memorial Hospital,Pawtucket, RI 02860.

6Author for correspondence.6Nonstandard abbreviations: CHD, coronary heart disease;

LDL-C, HDL-C, low- and high-density lipoprotein cholesterol,respectively; ape, apolipoprotein; CV,,, coefficient of analyticalvariation; and CV, coefficient of physiological variation.

Received January 8, 1991; accepted February 14, 1992.

proteins become more widely available, they undoubt-edly will also be used for clinical decision making.

Upon repeated measures, considerable variability oflipids and lipoproteins has been noted within individu-als (6-15). This variation arises in part from variationsin the accuracy and precision of the laboratory methodsused to measure these components (analytical varia-tion) but is also largely reflective of the normal biolog-ical fluctuations that occur in day-to-day living (physi-ological variation).

Criteria for accuracy and precision for the total cho-lesterol measurement have been defined to identifyproperly those individuals with an unacceptably highcholesterol concentration (16). The error attributable tothe analytical component of variation has been found tocontribute less than half the total variability of totalcholesterol measurements in normal individuals (16).Most of the intra-individual variation in total choles-terol is due to normal biological fluctuations (14, 16).

The National Cholesterol Education Program Labo-ratory Standardization Panel, assuming a coefficient ofanalytical variation (CV5) to be 5.0%, and the coefficientof physiological variation (CV) to be about 6.5%, esti-mated that duplicate measurements obtained in bloodsamples on at least four occasions would be necessary todetermine an individual’s “usual” cholesterol concentra-tion with 95% confIdence (16). For practical reasons,however, it was recommended that, as a minimum,single measurements made on at least two occasionscould be used for this purpose. This would suffice toestablish an individual’s usual cholesterol value with a93% certainty (16).

The information currently available about the normalphysiological variability of LDL-C and HDL-C (6,9-15)has been obtained for the most part in normal subjects.To our knowledge, the present study is the first toexamine physiological variation in a clinical setting inwhich the physician must establish the patient’s usuallipid and lipoprotein concentrations during the course ofevaluation and treatment. Here we report the analyticaland physiological variability of lipids, lipoprotein cho-lesterol, and apoA-I and apoB in patients referred forevaluation and treatment of dyslipidemia.

Materials and MethodsStudy Subjects

Subjects were selected from among patients of theJohns Hopkins University Lipid Referral Clinic, a ter-tiary-care referral clinic specializing in the detectionand management of lipid and lipoprotein disorders.Most patients referred to the clinic had seen a dietitianbefore their referral and were consuming a diet low in

CLINICALCHEMISTRY,Vol.38, No.6, 1992 865

saturated fat and cholesterol. The subjects were seen ina standardized format for the first three visits: Theywere asked to fast for 12 h before each visit, and bloodwas obtained for the measurement of lipids, lipoproteincholesterol, and apoA-I and apoB. A history and physi-cal examination and nutritional record were obtained atthe first visit. At the second visit (one month later), thesubject’s usual nutritional intake was reviewed by adietitian to ensure that the patient was following anAmerican Heart Association Step I diet. Any subjectwhose dietary fat intake exceeded that recommended bythe Step I diet was excluded from this study. Six weeksafter the second visit, patients returned to the clinic fora second review of their dietary intake and furtherconsultation with the physician. Medication usage andgeneral health were recorded in the medical record.

For the present study, we reviewed the medical rec-ords of all patients returning to the clinic for their thirdvisit between March 1, 1988, and December 31, 1989.Subjects selected for study must have completed allthree visits between January 1, 1988, and December 31,1989. Of 188 such patients, 26 were excluded becausethey were taking lipid-lowering medications, 10 wereexcluded because their dietary fat intake changed dur-ing the first three visits, and 1 patient had developedhypopituitarism. This left an initial study population of151 patients. For the purposes of this study, we calcu-lated LDL-C from the measured values of total choles-terol, triglycerides, and HDL-C as described by Friede-wald et al. (17) (see below). Because this equation is notused when triglycerides exceed 4.52 mmoIJL, we elimi-nated another 21 subjects whose triglycerides exceeded4.52 mmol/L at any visit and 2 subjects for whichHDL-C values were missing for one of their three visits.The final study population for total cholesterol, triglyc-erides, LDL-C, and HDL-C consisted of 128 patients whohad a complete lipoprotein profile at all three visits. Ofthe initial study population of 151 patients, 143 subjectshad results for apoA-I and apoB analyses at each of thethree visits. For the apolipoproteins, we did not excludesubjects with triglycerides exceeding 4.52 mmol/L.

Lipid andLipoproteinMeasurementsBlood was drawn into evacuated collection tubes con-

taining disodium EDTA, 1.5 gIL. Plasma was preparedwithin -2 h of venipuncture and the HDL-contaiingfraction was prepared within the following 2 h. Thesamples were stored at 4 #{176}Cfor one to three days beforeanalysis.

Cholesterol was measured enzymatically with a com-mercially available reagent (Cholesterol, High Perfor-mance; Boehringer-Mannheim Diagnostics, Indianapo-lis, IN; cat. no. 704121). Triglycerides were measuredenzymatically with the reagent available from AbbottLabs, Inc. (Abbott Park, IL; A-Gent Triglycerides).HDL-C was measured in the clear supernatant obtainedafter precipitation of apoB-containing lipoproteins withheparin and manganese chloride as described previ-ously (18). LDL-C was calculated from the empiricalrelationship of Friedewald et a!. (17): ELDL-C] = [total

chol] - [HDL-C] - [triglycerides]/5, where all concen-trations were expressed in g/L. We converted the lipidand lipoprotein cholesterol concentrations to mmol/L

and report them as such in this paper. During the periodof the study, the laboratory was standardized for choles-terol, triglycerides, and HDL-C measurements accord-ing to the criteria of the Centers for Disease Control-National Heart, Lung and Blood Institute Lipid Stan-dardization Program (19). ApoA-I and apoB weremeasured by radial immunodiffusion with commerciallyavailable immunodiffusion plates (apoA-I from Tago,Burlingame, CA; apoB as M-Partigen Apolipoprotein-Bfrom Behring Diagnostics, Somerville, NJ).

Calculation of CV5

The coefficients of analytical variation (CV,, values)for total cholesterol, triglycerides, and HDL-C weredetermined from analyses of several serum control poolsfurnished by the Centers for Disease Control (Atlanta,GA). Pools Q11, Q15, and Q18 were used for cholesteroland triglycerides, and AQ11 and AQ12 were used forHDL-C. Each pool was analyzed in duplicate with eachanalytical run. For apoA-I and apoB analyses, we usedtwo lyophilized pools (pools A and B) obtained commer-cially (Omega; Technicon Instruments, Tarrytown, NY)and several locally prepared frozen pools (pools 4,6,7,8,and 9). Not all the pools were analyzed in all runsduring the 22 months of the study, because new poolswere introduced as older pools were exhausted. For eachpoo1 we determined the mean and SD, which we thenused to calculate the CVa:

CV5 = (SD/mean) x 100

We found that for a particular analyte the CV,, valuesfor different quality-control pools were similar (see be-low). We therefore used the average CVa for the pools asthe CV,, for that analyte.

The CVa for LDL-C was estimated from the individualvariances for total cholesterol, triglycerides, andHDL-C, because HDL-C was not analyzed in the poolsused to measure total cholesterol and triglycerides, sowe did not have LDL measurements for the quality-control pools.

The CVa for LDL-C was calculated as follows, assum-ing that the quantifications of total cholesterol, HDL-C,and triglycerides (TG) were independent measurements:From the Friedewald equation,

[LDL-C] = [total chol] - [HDL-C] - [TG]/5.

Thus the analytical variance (Va) was as follows:

V5(j) = Va(total chol) + Va(irc) + Va(Tt/25

and CVa(LL) = [V5(LDL)]#{176}5/meanLDL

where mean LDL for the quality-control pools wascalculated from the mean pool values for total choles-

n

V2-V3

V1-V3

To provide insight into the range of CV for each Table 1. Mean (and SD) Age, Time between Visits (V),analyte, we determined the cumulative frequency of and Quetelet Index In Age- and Sex-Specific subgroupsaCV for all subjects combined. The median, 75th, and95th percentiles for the analyte in question were used

*ii su*cts Boys Men GIrls Women

128 42 27 28 31for certain calculations as indicated below. Age, years 28 9 51 9 52

For comparison, we also estimated CV by treating (24) (4) (12) (4) (17)the variability as variance in a random effects analysisof variance model as described by Fless (20). In thismethod, variance was partitioned into among-individ-ual and within-individual components. The among-in-dividual variability was treated as a random effect andwithin-individual variance was considered to consist ofbiological and analytical components. For all subjectscombined, the model included terms for male-femalevariability, age-group variability, interaction of age and

Between-visit interval,weeksV1V2 7.3 7.7 7.2

(5.1) (6.0) (6.2)12.7 13.3 11.2(9.2) (11.1) (6.4)

20.0 21.0 18.4(11.7) (14.4) (10.1)

QueteletIndexc)

sex, and individual within-sex and -age groups.

Comparison of Descriptive Characteristics

Within age categories, the mean ages of males andfemales were compared by use of Student’s t-test. Todetermine changes in weight across the thi visits,paired t-tests were performed for within-patient differ-

groups are defined as follows:boys, 1 8 years; men, >18 years;girls,alS years; women, >18 years.

ences in Quetelet Index between visits. Differences in bWeigf (kg)/[helght (rn))2.visit-to-visit time intervals across the four age-sex #{176}Valuesignificantly different (P = 0.01) fromV1 and V2.

6.5(2.6)14.3

(10.6)20.8

(11.1)

VI

7.4(4.8)11.8(6.9)19.2(9.7)

V2

V3

19.6 26.6(5.5) (3.2)19.5 26.7(5.6) (3.2)19.5 263C(6.0) (3.3)

18.1(3.8)18.1(3.7)18.2(4.0)

25.9(5.3)25.7(5.5)25.7(5.4)

866 CLINICALCHEMISTRY,Vol.38, No.6, 1992

terol, triglycerides, and HDL-C. The latter, in turn,were calculated by averaging the laboratory means foreach control pool for each measured variable. The ana-lytical variances for each of the measured variableswere calculated as

[CVa(average) X concentration(avee)]2

where CVa(,,ve) for a particular measured variablewas calculated from the values in Table 3 (see Results).

Calculation of CV,

For each analyte, the coefficient of physiological vari-ation, CVI,, was calculated as described by Warnick andAlbers (7). Briefly, we calculated for each subject themean and variance of the values for the three visits.This variance, V, represented the total variance for thesubject, including the contributions of physiological andanalytical variation:

V = V + Va

where V and Va are the physiological and analyticalvariances, respectively. Physiological variance was thencalculated from the relationship

V = V - Va

by substituting our estimates of V for each patient andthe average CV,, calculated as described above. CV andCV for each patient were calculated from the equations

CV [W)#{176}5/patientmean] X 100

CV = [(V)#{176}5/patientmean] x 100

groups were evaluated by analysis of variance. In addi-tion, Scheff#{233}’stest was used to test pair-wise compari-sons among groups.

All calculations were performed with the SAS pro-gram (21).

Results

The demographic characteristics are shown by age andsex in Table 1 for the 128 patients who had completelipoprotein profiles. The demographic characteristics ofthe group of 143 patients with all three apoB and apoA-Ivalues did not differ significantly from that presented inTable 1. Four age and sex categories were exsimined:males �18 years (n = 42); males >18 years (n = 27);females 18 years (n = 28); and females >18 years (n =

31).There were no significant differences (P >0.05) in the

mean age between boys and girls or between adult menand women. The intervals between visits were notsignificantly different for the four subgroups (P >0.05).The average interval between visits 1 and 3 was about20 weeks (Table 1). Adult men lost 1.1 (± 1.4) kgbetween visits 2 and 3 (P <0.01); this was secondary tocaloric restriction, because no change in the compositionof the diet occurred over the three visits. With thisexception, none of the subgroups significantly changedtheir body mass index (Quetelet Index) between visits.

The group mean values for cholesterol, triglycerides,LDL-C, HDL-C, and the two apolipoproteins are pre-sented in Table 2.

Analytical Variation

The CVa values for cholesterol, triglycerides, HDL-C,and the two apolipoproteins are shown in Table 3. The

GIrls Women

285.96

(1.39)1.43

(0.53)4.05(1.32)1.26

(0.31)

261.59

(0.37)1.40

(0.27)

nApoB

ApoA-I

317.15

(1.71)1.56

(0.67)4.78

(1.65)1.66

(0.54)

382.01

(0.50)1.66

(0.37)

1431.71

(0.46)1.45

(0.30)

431.48

(0.36)1.37

(0.25)

361.77

(0.41)1.34

(0.18)

Table 3. Cholesterol, TriglycerIdes, and HDL-C in Quality-Control Pools Analyzed during the Period of the StudyRef. Laboratory

Analyts Peel Period value’ n mean (SD) CV, S

Lipids and HDL-cholesterol, mmot’LChol 011 3/88-9/88 4.78 118 4.73 (0.09) 1.8

015 9/88-12/89 4.65 476 4.65(0.09) 2.0018 3/88-12/89 7.13 646 7.10 (0.15) 2.1

Trig 011 3/88-9/88 1.74 192 1.72(0.08) 5.0015 9/88-12/89 1.63 474 1.55(0.77) 5.0018 3/88-12/89 2.92 696 2.87(0.14) 4.8

HDL-C AOl 1 3/88-11/88 1.01 296 0.96(0.08) 8.4AQ12 10/88-12/89 1.34 496 1.39(0.08) 6.0

Apoilpoproteins, giLApoA-i Ab 3/88-6/88 74 1.742(0.097) 5.6

B 7/88-12/89 556 1.795(0.121) 6.7

4 3/88-6/88 40 1.418(0.056) 4.06 6/88-9/88 102 1.411 (0.094) 6.77 9/88-5/89 228 1.472 (0.104) 7.08 3/89-9/89 162 1.593 (0.100) 6.39 8/89-12./89 147 1.618(0.105) 6.5C 3/88-6/88 118 0.739 (0.037) 5.0B 7/88-12/89 830 0.822 (0.047) 5.76 4/88-9/88 184 1.808 (0.129) 7.17 9/88-5/89 352 1.324 (0.102) 7.78 3/89-9/89 243 1.310(0.093) 7.19 7/88-12/89 197 1.263 (0.074) 5.8

‘Reference values established by the Centers for Disease Control.Lettered pools represent lots ofOmega lyophilized control pools. Numbered pools were prepared Inthe laboratoryand stored at -70 ‘C until analyzed (see

MateriaLsand Methods).

ApoB

CLINICALCHEMISTRY, Vol. 38, No. 6, 1992 867

Table 2. Mean (and SD) Lipid and LipoproteinCholesterol Concentrations In Age- and Sex-Specific

Subgroups of Patients Referred to Johns HopkinsLipid Referral Cilnic

Analyte All subl.cts Boys Men

Lipids and lipoproteins, mmol/Lbn 128 42 27Total cholesterol 6.15 5.62 6.01

(1.39) (1.04) (0.80)Triglycerides 1.46 1.19 1.81

(0.60) (0.48) (0.55)LDL-C#{176} 4.12 3.76 4.01

(1.28) (1.03) (0.76)HDL-C 1.38 1.32 1.17

(0.40) (0.30) (0.21)Apoilpoproteins, gIL

For each analyte, we averaged the three values for each patient Theoverall mean (SD) values shown were calculated from the patient means.

bTo conveit values to mg/dL multiply cholesterol or Ilpoprotein-cholesterolvalues by 38.7 and triglycerIde values by 88.5.

#{176}Estlmated(see text).

serum control pools used for the measurement of totalcholesterol had reference values of 4.65, 4.78, and 7.13mmolJL, and the CVa for total cholesterol ranged from

1.8% to 2.1% in these pools. The reference values for the

triglyceride concentrations in these control pools were1.74, 1.63, and 2.92 mmol/L, and the CVa for triglycer-ides ranged from 4.8% to 5.0%. For HDL-C, the refer-ence values for the serum control pools were 1.01 and

1.34 mmolJL; the CVa was 8.4% and 6.0%, respectively.On the basis of these data, we calculated the followingaverage values for CVa: total cholesterol, 2.0%; triglyc-erides, 5.0%; HDL-C, 7.2%. These values were taken asthe analytical components of variation in the patients,because the lipid and lipoprotein concentrations of thepools were similar to the patients’ sample means (Table2). Based on the average CV,, for the three analytes, theCVa for LDL-C was 4.3%.

Two lyophilized and five frozen pools were analyzedfor the apolipoproteins. For apoA-I, CV, ranged from4.0% to 7.0%, and for apoB the range of CV,, was 5.0% to7.7% (Table 3). The CV,, for the lyophilized and frozenpools was similar for both apolipoproteins. On the basisof these results, we used average values of 6.1% (apoA-I)

and 6.4% (apoB) for the CV,, for the two apolipoproteins.

Physiological Variation

The median total (CV) and physiological (CV) coef-ficients of variation are presented in Table 4 for theoverall group and for the age- and sex-specific sub-groups. There were no significant differences in the totalor physiological CVs for total cholesterol, HDL-C,LDL-C, or apoA-I among any of the subgroups studied.In particular, no increase in the variability of these

WomenCVCV,

ever, CV,, might be expected to vary among individuals,the net effect of which is that a single measurement in apatient with a low CV more reliably estimates thatpatient’s “usual” concentration than does a single mea-

TC TO HDL-C LDL-C ApOA-l ApoB surernent in a patient with a high CV,,. To explore thisquestion further, we determined the cumulative fre-

CV,, 5.0 17.8 7.1 7.8 7.1 6.4 quencies of CV,, for each of the lipid, lipoprotein choles-CV, 5.3 19.2 9.6 8.8 10.1 8.2 terol, and apolipoprotein concentrations for each of the

four age-sex subgroups (Figure 1). The slopes of theCV,, 5.2 16.3 6.9 7.7 5.9 curves were similar between the four subgroups for anyCV, 5.5 19.1 10.2 9.1 9.8 8.1 analyte. For all of the a.nalytes, the distributions of CV

Men were similar for all four subgroups. This was particu-CV,, 5.2 17.3 7.7 7.0 5.7 7.4 larly true for total cholesterol and LDL-C. Table 5 showsCV, 5.6 18.0 11.0 8.2 9.1 8.3 the median, 75th percentile, 95th percentile, and ranges

Girls of CV,, for each of the four subgroups. As was reflected inCV,, 5.9 23.2 8.3 7.6 8.8 9.6 Table 4, the range of physiological CV was the least forCV, 6.2 24.7 10.1 8.5 11.8 11.0 total cholesterol and the greatest for triglycerides (Table

5). Also, the distributions of CV,, for LDL-C and apoB,and for HDL-C and apoA-I were superimposable foreach of the subgroups (Figure 1). This might be ex-pected, because apoB is found primarily in LDL andapoA-I in HDL.

CV,, was also estimated by using a random-effectsanalysis of variance (20). The CV,, values estimated inthis way ranged from approximately the 65th to 75thpercentiles of the frequency distributions for the various

analytes (Table 6).

868 CLINICALCHEMISTRY,Vol. 38, No. 6, 1992

Table 4. Median Total (CVJ and PhysIological (CV)CoefficIents of VariatIon (%)for All Subjects and for

Age- and Sex-SpecIfIc Subgroupsa for Lipids,Lipoprotelne, and Apolipoprotelns

SubJscts

All subjects

4.6 17.3 7.1 8.0 7.4 5.34.8 18.7 8.6 8.6 9.5 6.7

Tables 4-9: TC, total cholesterol; TO, triglycerides.‘CV, values used forthe calculations: TC, 2.0%; TO, 5.0%; HDL-C,7.2%;

WI-C, 4.3%; apoA-l, 6.1%; apoB, 6.4%.bFor TC, TO, HDL-C and WI-C, n = 128;forapoA-l and apoB, n = 143.

components was observed in children as compared withadults.

The CV for triglycerides in girls was higher than inwomen (P <0.02) (Wilcoxon rank-sum test). The varia-tion in total cholesterol and triglycerides over the threevisits was predominantly due to physiological variation.The median CV exceeded the CV,, by more than twofoldfor total cholesterol, and by more than threefold fortriglycerides. This pattern did not differ significantlyamong any of the subgroups. The mean CV was 6.5%and 2 1.2% for total cholesterol and triglycerides, respec-tively. Therefore, a subject with total cholesterol of 6.20mmol/L would be expected to have a measured valuebetween 5.80 and 6.60 mmol/L two-thirds of the time,and between 5.39 and 7.00 inmol/L 95% of the time, withmost of the variation between visits being due to phys-iological fluctuations attributable to factors of day-to-day living.

The median CV for HDL-C for the overall group was9.6%. Median CV (7.1%) was equal to CV,, (7.2%). Thus,the analytical component of variation contributed sub-stantially to the overall variability of HDL-C measure-ments. The median CV and CV values for the LDL-Cconcentration in the overall group were 8.8% and 7.8%,respectively. The CV for LDL-C exceeded the CV,,(4.3%) by about 1.8-fold.

The median CV and CV,, values for apoA-I were10.1% and 7.1%, respectively; those for apoB were 8.2%and 6.4%, respectively. The physiological fluctuation inapoA-I was therefore equal to that in HDL-C and thatfor apoB was a little less than for LDL-C (Table 4).

Cumulative Frequencies of CV

The CV values shown in Table 4 are the medians forthe patients and patient subgroups as a whole. How-

DiscussionThe analytical and physiological components of total

variation were calculated for total cholesterol, triglyc-erides, HDL-C, LDL-C, and apoA-I and apoB in hyper-lipidemic subjects. Most of the patients were physician-referred and had already made an attempt to modifytheir saturated fat intake before their initial visit. Nosubstantial changes in dietary fat intake were reportedbetween the first three visits in any of the subjects;however, most subjects reported more conscientious eat-ing habits after the first visit. This could explain, inpart, the trend towards more desirable lipid, lipoproteincholesterol, and apolipoprotein concentrations betweenvisits one and three in the overall group. Mean totalcholesterol concentration decreased 3.1% between visitsone and three. During this period, triglycerides do-creased 8.0%, apoA-I increased 4.6%, and apoB de-creased 2.7% (P <0.05). There were no significantchanges in HDL-C or LDL-C. In view of these observa-tions, we think it likely that part of the observedvariability was due to self-intervention as the patientsbecame more aware of their eating habits with eachsucceeding visit. However, this is likely to occur in mostclinical practices, and the variation observed in ourstudy seems particularly relevant to general medicalpractice settings.

The total variation in lipids and lipoproteins analyzedin serial samples from a given individual is generallypartitioned into two components. The analytical compo-nent of variation generally includes pre-ana.lytical fac-tors such as patient preparation, blood drawing, andsample preparation as well as factors related to the

100

80

60

100

80

60

40

20

0

120

100

80

60

40

20

0

40

20

Triglycerides

0 5 10 15 20 25 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30

120

100

120

80

60

40

20

0 10 20 30 40 50 60 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35

Physiological CV (%)

arise from a variety of sources that affect lipoprotein Table 5. Percentiles and Ranges for CV,, Age and Sexmetabolism, including diet, exercise, smoking, use ofcertain medications, alcohol, and the patient’s generalstate of health. Seasonal variations (23-25) and pos-tural changes (26-28) also affect lipid and lipoproteinconcentrations.

The CVs we found for total cholesterol, triglycerides,HDL-C, and LDL-C are similar to those found in severalother laboratories (6, 7). The CVa we observed for totalcholesterol and triglycerides-2.0% and 5.0%, respec-

All subjectsMedian75%lIe95%ileRange

BOYSMedian

TC TG HDL.C LDL-C ApoA4 ApeS

5.09.0

14.00-24.7

5.2

17.826.343.6

0-57.3

16.3

CV, ‘‘

7.1 7.813.7 11.524.5 20.0

0-28.5 0-36.0

6.9 7.7

7.112.020.1

0-32.0

5.9

6.412.019.6

0-39.5

5.7tively-are comparable with previous reports of 1-3% 75%ilo 9.1 26.9 13.4 11.4 11.5 14.5for total cholesterol and 1.5-5% for triglycerides (7, 95%ile 15.2 45.2 24.7 26.2 20.1 20.210-15). The CV,, for HDL-C can be expected to exceedthat for total cholesterol or triglycerides because of the

RangeGirls

0-24.7 0-57.3 0-28.5 0-36.2 0-31.1 0-39.5

additional steps required to precipitate apoB-containing Median 5.9 23.2 8.3 7.6 8.8 9.6lipoprotein before measuring the cholesterol attribut- 75%lle 9.3 30.9 15.8 11.0 14.1 12.4able to the HDL fraction (29). We obtained a CV,, of 7.2% 95%lle 14.0 45.5 25.7 17.0 24.7 16.3

for HDL-C, a value intermediate between those reported Range 0-18.0 0-46.7 0-26.8 0-21.6 0-27.0 0-22.1

in the Lipid Research Clinics Program (30,31) and those Men

reported in several other studies (10-12, 14, 15). Median 5.2 17.3 7.7 7.0 5.7 7.4The CV for LDL-C calculated from the Friedewald

equation included the analytical variations of the totalcholesterol, triglycerides, and HDL-C measurements.Friedlander et al. (6) reported an overall CV of about13.0% for LDL-C but did not report the analytical andphysiological components separately. Siekmeier et al.(32) reported CV,, values of 6.0-12.0%, using variousmethods for precipitation of the apoB-containing lipo-

75%ile95%ileRange

WomenMedian75%IIe95%ileRange

8.211.7

0-14.9

4.66.6

12.50-14.2

22.639.6

0-41.7

17.321.843.6

0-45.1

14.1 12.119.2 17.0

0-23.3 0-18.0

7.1 8.012.1 11.124.5 20.0

0-25.2 0-21.1

10.220.6

0-21.3

7.413.323.8

0-32.0

12.126.0

0-31.0

5.38.7

12.40-23.5

120 Total Cholesterol 120 LOL-cholesterol 120 HOL-cholesterol

CLINICALCHEMISTRY,Vol.38, No.6, 1992 869

>

0

a)

cra)

U-

ci)>

-4-

0

E0

Fig. 1. CumulatIvefrequency(%) of CV,, (%) for totalcholesterol,LDL-C, HDL-C, triglycerides,apoA-I, and apoBThe distribution of CV,,values for each analyte was determined forthe four age-sex subgroups: men (#{149}),women (0), boys (V), and girls (V)

Table 7. Total Coefficient of VarIation (%) for Mean Analyte Value as a Function of the Number of Samples perPatient and ReplIcateAnalysesper Sample

No. of spsclm.n1 No. of .pecIm.ns

2 3 4 1Analyt. CV,, CV. 1 3 4

TC 5.0 2.0 5.4 4.1 35 3.2 5.2 3.8 3.2 2.9TG 17.8 5.0 18.5 13.5 11.4 10.2 18.2 13.1 10.9 9.6LDL-C 7.8 4.3 8.9 7.0 6.2 5.8 8.4 6.3 5.4 4.9HDL-C 7.1 7.2 10.1 8.8 8.3 8.0 8.7 7.2 6.5 6.2ApoA-l 7.1 6.1 9.4 7.9 73 7.1 8.3 6.6 6.0 5.6ApoB 6.4 6.4 9.1 7.8 7A 7.2 7.8 6.4 5.8 5.5

2

#{149}One analysls/epeclmen.bTwo analyses/specimen.CTh physiologiCal coefficient of variations are the median values for all subjects (from Table 4).

870 CLINICAL CHEMISTRY, Vol. 38, No. 6, 1992

Table 6. CV,, Calculated from Random Effects Analysisof Variance

TC TO HDL.C LDL-C Ap0A-l ApoB

CV,,, S

All subjects 7.4 25.3 10.4 10.4 9.2 9.7Boys 7.8 23.4 11.3 11.0 11.1 11.2Girls 7.7 30.5 11.2 9.5 11.4 10.5Mon 6.9 23.5 10.1 9.5 9.0 10.0Women 7.0 23.1 9.0 10.8 10.2 7.8

proteins. The CV,, we observed for LDL-C (4.3%) wassomewhat lower, and closer to the values (4.4-5.2%)reported by others (11, 14, 15).

The median physiological variabilities of all the ana-lytes except plasma triglycerides were similar. CV,,ranged from 5.0% for total cholesterol to 7.8% for LDL-C

and was in the ranges reported by others for free-livingnormal populations (6, 7, 11, 12, 14, 15). Those forapoA-I and apoB were similar at 7.1% and 6.4%, respec-tively, similar to the values reported by Brown et al. (11)but somewhat larger than values reported by Ford (12).The CV for triglycerides (17.8%) was considerablyhigher tcian that of the other analytes, as has been

reported by others (6, 7, 11, 12, 14, 15), but lower thanvalues of >20% reported in several studies of normalpopulations (11, 14, 15).

In another approach to the estimation of CV,,, thevariability was treated as variance in a random-effectsanalysis of variance (20). With this method, the within-subject variance was partitioned into analytical andphysiological variances. This method gave a summaryestimate for CV,, that was somewhat higher than the50th percentile values obtained when the CV,, wasestimated for each individual than when summarized

across the entire group: The individual variances wereskewed to the right, whereas the individual means werenot. By dividing the individual variances by the indi-vidual means before summarizing across individuals,the effects of the high variances on the CV,, estimateswere diminished, resulting in lower overall estimates ofCV,,. Although the single CV,, as derived from therandom-effects analysis of variance best describes thestudy population as a whole, the calculation of individ-ual CV,, values allowed us to examine how these were

distributed in the study group and to calculate CV forthe means of measurements in serial samples at variouspercentiles for CV,,.

The extent to which the lipids, lipoprotein, and ape-lipoprotein concentrations vary over time is clinicallyrelevant for the physician who must determine both thepatient’s risk for cardiovascular disease attributable todyslipidemia and the response to dietary and drugtreatments. In view of the variability of lipids, lipopro-teins, and apolipoproteins within individuals, severalsamples obtained at intervals of at least several weeksare required to establish the patient’s usual concentra-tions of these components. Table 7 illustrates the totalcoefficient of variation (CVi) for the mean value of eachanalyte as a function of the number of occasions thepatient is sampled when the samples are analyzed onceor in duplicate. The values in Table 7 were calculatedfrom the median CV,, and CV,, values we observed inthis study. CV for total cholesterol was 5.4% whenmeasured once in a single sample and decreased to 5.2%if the analysis was performed in duplicate. In contrast,CV decreased to 4.1% if two serial samples were takenfrom the patient and each analyzed once. Thus, more isgained by obtaining two different samples from thepatient than by analyzing a single sample in duplicate.As a matter of practicality, an individual sample isunlikely to be analyzed in replicate in the routineclinical laboratory, and the question arises of how manysamples are required to establish the patient’s “usual”or average concentration of the component of interest.The National Cholesterol Education Program Labora-tory Standardization Panel recommended that the pa-tient be sampled for cholesterol measurement on atleast two different occasions at least one week apart(16). As seen in Table 7, such a sampling schemereduces the CV for total cholesterol to 4.1%. Table 7also reveals, however, that duplicate analyses of three

or four separate specimens would be required to reducethe CV, for LDL-C, HDL-C, apoA-I, and apoB concen-trations to about 5%, and for triglyceride concentrationto 8%. Such extensive analyses would not be practical,for economic and other constraints. On the basis of ourfindings, however, we recommend that at least threesamples be obtained from the patient, each to be ana-lyzed once. Under these conditions, triglycerides would

BoysGirlsMenWomen

CLINICALCHEMISTRY, Vol.38, No. 6, 1992 871

be established with a CV of about 11%; LDL-C, HDL-C,apoA-I, and apoB with CV values of 6% to 8%; and totalcholesterol with a CV, of about 3.5% (Table 7, under-lined values).

The information given in Table 7 is useful for indicat-ing the number of serial samples required to provide anestimate of a patient’s usual concentrations on average.However, because the median CV,, values were used inthis table, these estimates would apply to only half ofthe patients in our study, and the number of serialsamples required would be greater for the rest of thepatients. To illustrate this, we used the 75th and 95thpercentiles for the CV,, distributions in Table 5 tocalculate the number of serial specimens, each analyzedonce, that would be required to estimate the patient’susual concentrations for each analyte with about thesame precision as provided by three serial specimens,each analyzed once, based on the median CV,, valuesshown in Table 7. The results (Table 8) indicate that forpatients at the 75th percentile for CV,,, from 6 to 15serial samples would be required, depending on agegroup, sex, and analyte; for patients at the 95th percen-tile, 12 to >20 serial specimens would be required.Again, analyzing this many baseline samples to estab-lish the patient’s usual lipid, lipoprotein or apolipopro-tein concentrations would be quite involved and ex-tremely expensive, and such a protocol would not beattempted in the usual clinical settings or even for mostresearch purposes.

Table 9 illustrates the total coefficients of variationfor single measurements in each of three serial samplesfor patients with CV,, values at the 75th percentilesshown in Table 5. For total cholesterol, which manifeststhe least physiological variation among the six ana-lytes, measurements in three serial samples would es-tablish the patient’s usual cholesterol concentrationwith an overall precision (CV,) of about 5.6% (Table 9),compared with 3.5% for a patient with a median CV,,(Table 7). Even for triglycerides, the most variable of thesix analytes, three serial samples would suffice to estab-lish the patient’s triglyceride concentration with anoverall precision of about 16.0% (Table 9), comparedwith 11.4% for patients with a median CV,, for triglyc-

Table 8. Serial Specimens Required to EstimatePatients’ “Usual” Lipid, Lipoprotein, and

Apolipoprotein Concentrations,a Based on 75th and95th Percentiles for CV,,

TC TO HDL-C LDL-C ApoA-l ApoB

75th 95th 75th 95th 75th 95th 75th 95th 75th 95th 75th 95th

10 >20 7 >20 11 >20 7 >20 8 >20 15 >2011 >20 9 20 15 >20 6 14 12 >20 11 198 16 5 15 12 20 7 15 7 >20 12 >205 19 5 18 9 >20 6 20 11 >20 6 12

#{149}Numberof specimens requiredto estimate usual concentration with aboutthe same precisions as provided by three serial samples, each analyzed once,for patients with median CV,,,as shown in Table 7 for assuming eachspecimen is analyzed once. CV values used for the calculations are thoseshown in Table 7.

Table 9. Total Coefficient of Variation (CVi, %) for MeanAnalyte Value in Patients wfth CV,, at the 75th

PercentlleIC TO HDL-C LDL-C ApoA-I ApeS

All subjects 5.6 16.0 10.7 7.9 9.2 9.4Boys 5.6 16.3 10.6 7.9 9.0 10.5Girls 5.7 18.5 11.6 7.8 10.2 9.6MenWomen

5.14.3

14.013.5

10.910.3

8.27.7

8.59.8

9.58.1

#{149}Assuming single measurements In each of three serial samples. The CV,values shown In Table 5 were used for these calculations.

erides (Table 7). Thus, the loss in precision for patientsat the 75th percentile for CV,, is relatively modest, andin most circumstances the effort and costs of makingmeasurements in more than three serial samples wouldprobably not be justified.

The biological variations measured in the presentstudy were determined from measurements made over aperiod of about five months in a group of patients whowere seen over a two-year period. The findings aretherefore probably representative of those to be ex-pected in routine clinical practice with patients who arebeing evaluated for hyperlipidemia. An understandingof the magnitude of the normal variations that occur inlipid, lipoprotein, and apolipoprotein concentrations isrequired for the proper interpretation of the laboratorydata during diagnosis and in evaluating the effective-ness of treatment.

This work was supported by the Johns Hopkins Lipid ResearchAtherosclerosis Unit. We thank Dr. Henry Feldman, Senior Re-search Scientist, New England Research Institute, Watertown,MA, for help with the statistical analyses.

References1. Castelli WP, Garrison RJ, Wilson PWF, Abbott RD, KalousdisnS, Kannel WB. Incidence of coronary heart disease and lipoproteinlevels. The Framingham study. JAmMed Assoc 1986;256:2835-8.2. Bachorik PS, Kwiterovich P0. Apolipoprotein measurementsin clinical biochemistry and their utility via-a-via conventionalassays. Clin Chim Acta 1988;178:1-34.3. Castelli WP, Doyle JT, Gordon T, et al. HDL cholesterol andother lipids in coronary heart disease-the co-operative lipoproteinphenotyping study. Circulation 1972;55:767-72.4. Gordon T, Castelli WP, Hjortland MC, Kannel WB, Dawber TR.High density lipoprotein as a protective factor against coronaryheart disease. The Framingham study. Am J Med 1977;62:707-14.5. National Cholesterol EducationProgramExpert Panel. Reportof the National Cholesterol Education Program Expert Panel onDetection, Evaluation and Treatment of High Blood Cholesterol inAdults. Arch Intern Med 1988;148:36-69.6. Friedlander Y, Kark JD, Stein Y. Variability of plasma lipidsand lipoproteins: the Jerusalem Lipid Research Clinic study. ClinChem 1985;31:1121-.6.7. Warnick GR,Albers JJ. Physiological and analytical variationin cholesterol and triglycerides. Lipids 1975;11:203-8.8. Howes LG, Krum H. Intra-individual variability of plasmacholesterol concentrations [Abstractj. Clin Chem 1988;34:1930-1.9. Chiba K, Watanabe T, Ikeda M. Variability of serum highdensity lipoprotein cholesterol concentration in healthy subjects ina three-year term. J Epidemiol Community Health 1984;38:195-7.10. Demacker PNM, Schade RWB, Jansen RTP, Van’t Laar A.Intra-individual variation of serum cholesterol, triglycerides andhigh density lipoprotein cholesterol in normal humans. Athero-sclerosis 1982;45:259-66.11. Brown SA, Boerwinkle E, Kashanian FK, Swanson N, Patach

872 CLINICALCHEMISTRY,Vol.38, No.6, 1992

W. Variation in concentrations of lipids, lipoprotein lipids, andapolipoproteins A-I and B in plasma from healthy women. ClinChem 1990;36:207-1O.12. Ford RP. Essential data derived from biological variation forestablishment and use of lipid analyses. Ann Clin Biochem 1989;26:281-5.13 Hoizel WGE. Influence of hypertension and antihypertensivedrugs on the biological intra individual variation of electrolytesand lipids in serum. Cliii Chem 1988;34:1485-8.14. Mogadenn M, Ahmed SW, Mensch AA, Godwin ID. Withinperson fluctuations of serum cholesterol and lipoproteins. ArchIntern Med 1990;150:1645-8.15. Bookstein L, Gidding 58, Donovan M, Smith FL Day-to-dayvariability of serum cholesterol, triglyceride and high densitylipoprotem cholesterol levels. Arch Intern Med 1990;150:1653-7.16. National Cholesterol Education Program Laboratory Stan-dardization Panel. Recommendations for improving cholesterolmeasurement. Bethesda, MD: NIH Pub!. No. 90-2964, 1990.17. Friedewald WT, Levy RI, Fredrickson DS. Estimation of theconcentration of low density lipoprotein cholesterol in plasmawithout use of the preparative ultracentrifuge. Clin Chem 1972;18:499-502.18. Bachorik PS, Walker RE, Virgil DG. High density lipoproteincholesterol in heparin-MnC12 supernates determined with theDow enzymatic method after precipitation of Mn2 with HCO3.Clin Chem 1984;30:839-42.19. Myers GL, Cooper GR, Wins CL, Smith SJ. The Centers forDisease Control-National Heart, Lung and Blood Institute LipidStandardization Program. An approach to accurate and preciselipid measurements. Cliii Lab Med 1989;9:105-35.20. Fleas JL. The design andanalysis of clinical experiments. NewYork: Wiley and Sons, 1986:1-15.21. SAS/STAT Guide for personal computers version, 6th ed.Cary, NC: SAS Institute, Inc., 1987.

22. Harris EK, Yasaka T. On the calculation of a “referencechange” for comparing two consecutive measurements. Clin Chem1983;29:25-30.23. Brown SA, Epps DF, Dunn JK, et al. Effect of blood collectionand processing on radioimmunoasaay results for apolipoprotein Bin plasma. Clin Chem 1990;36:1662-6.24. Gordon DJ, Hyde J, Treat DC, et a!. Cyclic seasonal variationin plasma lipid and lipoprotein levels: the Lipid Research ClinicsCoronary Primary Prevention Trial placebo group. J Clin Epide-miol 1988;41:679-89.25. Buxtorf JC, Baudet MF, Martin C, Richard JL, Jacotot B.Seasonal variations of serum lipids and apolipoproteuns. Ann NutrMetab 1988;32:68-74.26. Dannenberg AL, Keller JB, Wilson PWF, Castelli WP. Lei-sure time physical activity in the Framingham offspring study.Description, seasonal variation and risk factor correlates. Am JEpidemiol 1989;129:76-88.27. Stoker N, Wynn V, Robertson G. Effect of posture on theplasma cholesterol level. Br Med J 1966;i:336-8.28. Tan MI!, Wilmhurst EG, Gleason RE, Soeldner JS. Effect ofposture on serum lipids. N Engl J Med 1973;289:416-9.29. Dixon M, Paterson CR.Posture andthe composition of plasma.Clin Chem 1978;24:284-.6.30. Bachorik PS, Most B, Lippel K, et a!. Plasma lipoproteinanalysis: relative precision of total cholesterol and lipoproteuncholesterol measurements in 12 Lipid Research Clinics laborato-ries. Clin Chem 1981;27:1217-22.31. Albers JJ, Warnick RW,Johnson N, et a!. Quality control ofplasma high density lipoprotein cholesterol measurement meth-ods. The Lipid Research Clinics Program prevalence study. Circu.lation 1980;62(Suppl IV):9-18.32. Siekmeier R, Marz W, Gross W. Precipitation of LDL withsulfated polyaniona: three methods compared. Clin Chim Acta1988;177:221-30.