Calculated Values for Low-Density LipoproteinCholesterol in the Assessment of Lipid Abnormalities and Coronary Disease Risk

CLIN. CHEM. /1, - (199) Calculated Values for Low-Density LipoproteinCholesterol in the Assessment of Lipid Abnormalities and Coronary Disease Risk Judith R. McNamara, Jeffrey S. Cohn, Peter W. F. WIlson,and ErnstJ. Schaefer Low-density Ilpoprotein (LOL) cholesterol concentrations are most commonly estimated by the formula LDL cholesterol = total cholesterol - [triglycerides (TG)/ + high-density lipoprotein cholesterol], although alternative factors such as TG/ have also been used. Using standardized, automated, enzymatic lipid assays, we analyzed 797 plasma samples from normal and dyslipidemic adults, to compare LDL cholesterol concentrations obtained after ultracentrifugation with those calculated by several such methods (i.e., TG/-TG/). For TG concentrations O. g/l, TG/ agreed best with the direct assay; for TG of.1-. g/l, TG/. was best; and for TG of.1-. g/l, TG/ was best. Differences in estimated values were generally small, however. At TG >. g/l, none of the factors tested allowed a reliable estimate of LDL cholesterol. When TG were. gil, % of estimated LDL cholesterol values were properly classified according to National Cholesterol Education Program cutpoints when the factor TG/ was used. We conclude that a convenient direct method for measuring LDL cholesterol is needed but, until one is available, use of the factor TGI will assure that most individuals with TG. g/l, as measured in a standardized laboratory, can be reasonably well classified for risk of coronary artery disease. AdditIonal Keyphraees: estimation compared with direct assay (post-u!tracentrifugation). screening heart disease risk factors Several studies have shown a significantly increased risk of coronary artery disease (CAD) with increasing concentrations of cholesterol in serum or plasma. This increase in risk is primarily due to increased low-density lipoprotein (LDL) cholesterol (1-). Moreover, prospectivestudieshave shown that when high LDL cholesterol concentrations are decreased by use of diet and drugs, the subsequent incidence of CAD is diminished (-). Therefore, the National Cholesterol Education Program s (NCEP) Adult Treatment Panel has recommended the screening of adults for cholesterol abnormalities to assess each individual s CAD risk (9). When the total cholesterol concentration exceeds. Lipid Metabolism Laboratory, USDA Human Nutrition Research Center on Aging at Tufts University, 1 Washington St., Boston, MA 111. F thngham Heart Study, National Heart, Blood and Lung Institute, Framingham, MA. Address correspondence to this author. Presented in part at ClinChem-, Springfield, MA, November, 19. Nonstandard abbreviations: VLDL, LDL, and HDL, very-low, low- and high-density lipoproteins; TG, triglycerides; CAD, coronary artery disease; HNRC, USDA Human Nutrition Research Center on Aging at Tufts University; FHS, Framingham Heart Study; NCEP, National Cholesterol Education Program; CDC, Centers for Disease Control; and NHLBI, National Heart, Lung, and Blood Institute, NIH. Received June, 199; accepted August 1, 199. g/l, or when it exceeds. g/l and two or more other CAD risk factors are present, NCEP has recommended further lipoprotein testing to determine the concentration of LDL cholesterol. LDL cholesterol concentration then becomes the basis for definitive dietary and drug treatment. Currently, LDL cholesterol concentration cannot be determined directly from serum or plasma without use of cumbersome ultracentrifugation procedures. The reference procedure for measuring LDL cholesterol is based on measurement of cholesterol in the density fraction >1. kgfl after ultracentrifugation (1). Because ultracentrifugation is not available in most clinical laboratories, assessment of LDL cholesterol usually is based on indirect calculations made from total cholesterol, high-density lipoprotein (HDL) cholesterol, and triglycerides (TG), which are usually unblanked for free glycerol (11, 1). The validity of calculated LDL cholesterol values therefore does not depend on the accuracy of one direct assay, but rather on the accuracy of three other assays, plus a mathematical calculation factor that estimates the amount of cholesterol in very-low-density lipoproteins (VLDL). We wanted to test the effect of using calculation factors on the categorizationof subjects in a lipid clinic and in a normal population, with respect to NCEP guidelines. Using assays standardized by the Centers for Disease Control- National Heart, Lung, and Blood Institute (CDC-NHLBI) (), we compared LDL cholesterol values obtained after ultracentrifugation (1) with those obtained by estimation. We tested several estimation factors: (a) the formula of Friedewald et al. (11), (b) the method of DeLong et al. (1), and (c) other similar factors to see which of these might give more nearly accurate estimations of LDL cholesterol at various concentrations of TG. Materials and Methods Study Populations The two independent populations of adults, all older than years,that were used for the present analysis had blood sampled after a 1- to 1-h overnight fast. The first population of 19 individuals consisted of two major subgroups: dyslipidemic patients referred to the Lipid Clinic at New England Medical Center, Boston, MA, and 7 research study subjects from the USDA Human Nutrition Research Center on Aging at Tufts University (HNRC). Plasma lipids and lipoproteins for both subgroups were determined by the HNRC Lipid Metabolism Laboratory. The second population consisted of adults who participated in Cycle of the offspring follow-up of the Framingham Heart Study (FHS), Framinghain, MA, for whom lipid and lipoprotein analyses were performed in the FHS laboratory. The FHS population represented an epidemiological cross-section of a normally distributedpopulation (mean ± SD: TG = 1. ± 1.1 g/l, LDL cholesterol = 1. ±. g/l). The HNRC population, however, consisted of two CLINICAL CHEMISTRY, Vol., No. 1, 199

diverse subgroups: the Lipid Clinic patients had a high predominance of lipid abnormalities (TG =. ±.1 g/l, LDL cholesterol = 1. ±. g/l), and the research subjects generally had low-to-normal lipid values (TG =. ±.1 g/l, LDL cholesterol = 1. ±. g/l). The only samples excluded before this data set (n = 797) was compiled were samples drawn from non-fasting individuals, those for which ultracentrifugation was not performed, and those from individuals younger than 1 years. Upoprotein Analysis Blood was drawn into tubes containingedta (1. gil final concentration). Plasma, separated after centrifugation (1 X g, mm, #{17}C), was stored at #{17}C until analysis. For both populations, lipid and lipoprotein analyses were performed by identical methodogy, but in separate laboratory facilities. An aliquot of each sample was subjectedtoultracentrifugation (1 h, x g, #{17}C) at a plasma density of 1. kgfl, with L- ultracentrifuges and.ti rotors (Beckman Instruments, Inc., Palo Alto, CA). After ultracentrifugation, the d <1. fraction was separated from the d >1. fraction by tube slicing (Nuclear Supply & Service Co., Washington, DC). Total cholesterol, TG (unblanked for free glycerol), 1. kg/l infranate cholesterol, and HDL cholesterol concentrations were determined with Abbott Diagnostics ABA- bichromatic analyzers and enzymatic reagents (Abbott s A- GENT ), as previously described (), after precipitation with dextran-mg (1). Lipoprotein cholesterol was calculated by subtraction: total cholesterol - infranate cholesterol = VLDL cholesterol; infranate cholesterol - HDL cholesterol = LDL cholesterol. All HNRC lipid analyses were standardized through the Lipid Standardization Program of the CDC-NHLBI. In addition, LDL cholesterol concentrations were estimated by seven different calculation formulas: (a) the method of Friedewald et al. (11), where total cholesterol - (TG/ + HDL cholesterol) = LDL cholesterol; (b) the method of DeLong et al. (1) with the same formula, except that TG/ is used in place of TG/; and (c-g) by use of the same formula, but with TG/, TGI., TG/., TG/7, or TG/. Statistical Analysis The variables utilized in this study were entered and stored in a VAX 11/7 computer (Digital Equipment Co., Maynard, MA) by using the scientific package RS/i (BBN Research Systems).Mathematical calculations were made with the RB/i software. Paired t-test analyses, performed with the Statistical Analysis Systems (SAS) software (SAS Institute, Cary, NC), were used to compare the significance of the differences between the various calculated estimations of LDL cholesterol and the concentrations determined after ultracentrifligation. Results After ultracentrifugation, the LDL cholesterol concentrations measured in the 19 HNRC samples and the FHS samples were compared with values calculated by each of the estimation methods described above. Because the results for each population, based on unblanked PG concentration and NCEP classification, were very similar, the data were combined. The values shown in Table 1 representthe percentage of samples in each unbianked PG range in which the LDL cholesterol concentration by estimation differed by <1% from the measured concentration. When TG concentrations were s. gil, use of the factor TG/ led to the highest percentage of individuals having values for estimated LDL cholesterol that fell within 1% of measured LDL cholesterol (7%, vs % for TG/. and % for TG/). The factor TG/. gave the closest results, 9%, when PG concentrations were.1-. g/l, vs % for TG/ and % for TG/. However, the differences in the estimated LDL cholesterol values between these three factors were generally small at PG concentrations s.#{17} g/l. Use of the factor TG/ (Friedewald) gave the closest estimations when TG concentrations were.1-. g/l: 7%, vs % for TGI, 9% for TG/., and 7% for TG/.. Above. g/l, no single best estimation factor emerged, but use of the factors TG/ to PG/ generally yieldedthe highest percentages. In the TG range of.1-. g/l, PG/ had the highest percentage of values within 1% of the measured values, but that encompassed only % of the samples with TG values in this range. Additionally, only factors TG/. and TG/ had estimated values that were not significantly different from measured values in the same Table 1. Percent of Samples for Which Estimated LDL Cholesterol Was within 1% of Measured LDL Cholesterol Factor used In estimation TG, g/l n TG/ TG/. TG/ TG/. TG/ TGI7 TGI O..1-1. 1.1-1. 1.1-..1-. 19 11 9 7*b 79 9* 7*b *b 7 77* 7 7 1 79 7 9 77 7 9 7 9.1-. 7.1-. 9 19 b 1 * 1 1.1-. 1 b 1b * 1.1-1. 1 iga 1 1.1-1. 7 1b l* 7b >1. b 9*b a Withineach TG range, a indicatestheestimationfactor with the greatestnumber ofsamples with less than 1% difference betweenmeasured andestimated LIX cholesterol concentration. b Estimation factors where the LDL cholesterol concentration was not significantly different(p >.) from the measured concentration by paired ftest analysis. CLINICALCHEMISTRY, Vol., No. 1, 199 7

PG range by paired t-test analysis. At concentrations of.1-. gil, PG/ and TG/. gave the closest agreement (%); at concentrations of.1-1. g/l, TG/ was closest at 7%; the 1.1-1. g/l range had the best agreement when estimation was calculated with the factor PG/ (1%); and forconcentrations >1. g/l, PG/ had the highest percentage of agreement (9%). Similar results have previously been reported by Kuba and Frantz (1). When the data were analyzed for percent error in the estimated values, similar trends were noted (Table and Figure 1). PG concentrations <. g/l had the lowest percent error when TG/ was used. The percent errorwhen PG were in the range of.1-1. g/l was evenly split between TG/ and TG/.. Concentrations of 1.1-. g/l also had the lowest percent error with TG/.. When PG values were.1-. g/l, LDL cholesterol concentrations had the least error when TG/ was used. When they were >. gil, PG1 generally continued to give the smallest mean percent error, but with increasingly large standard deviations. Variability started to become noticeable when PG concentrations were >. gil, and when PG concentrations were.1-. g/l at least 1% of individuals had estimated LDL cholesterol concentrations that deviated from the measured values by >1%. At TG concentrations >. g/l this percentage increased to more than 7% of individuals (Table 1). As shown in Table and Figure 1, the potential for significant estimation errors increased steadily with the PG concentration. We then subdividedthe data (using NCEP criteria for total cholesterol and PG) to separately evaluate normolipidemics (cholesterol <., TG. gil, n = 9), hypercholesterolemics (cholesterol., TG. g/l, n = 9), moderate hypertriglyceridemics (cholesterol <., PG.1-. g/l, n = 17), and combined hyperlipidemics (cholesterol., PG.1-. g/l, n = ). In addition, a subset of the combined hyperlipidemics with VLDL cholesterol/pg ratios. (n = ) were analyzed separately as possible Type III hyperlipoproteinemics. Severe hypertriglyceridemics (n = 7) were not analyzed in these subgroups. The data are summarized in Table. The accuracy of estimation for the normolipidemics and the hypercholesterolemics was similar to that for the entire group shown in Table 1. The best estimating factor for the normolipidemics was TG/., with % having estimated UI I.- w - 1. 11-1. 1.1-..1-..1-.1-..1-.,1-1. TRLYCERIDE CONCENTRATION (gel) Fig. 1. Percent error for estimated LDL cholesterol associatedwith seven estimationfactors(tg/, TG/., TG/, TG/., TGI, TG/7, TGI) at varioustg ranges(gil) Each bar representsone standard deviationfromthe mean percent errorfor each factor. The dotted linerepresentsthe 1% cutpoint used for evaluating acceptable error.the #{19} shows the bar(s) in each TG rangewiththe smallest percent error LDL cholesterol values within 1% of the measured values. Normolipidemics, as defined here, represented 7% of our entire population. Hypercholesterolemics had an estimation accuracy rate of 91% with the factor PG/ and represented 19% of the population. Hypertriglyceridemics with PG.1-. gil were best estimated with factors TG/. and TG/, but only correctly estimated % of samples. The hypertriglyceridemics with PG >. gil were best estimated with the factor TG/; however, only 7% were correctly estimated within the 1% limits. Combined hyperlipidemics with PG values of.1-. gil were correctly estimated in 7% of cases using PG/.; for those with PG.1-. gil, % were correctly estimated with the factor TG/. The subgroup of combined hyperlipidem- ics, who, in addition to above-normal cholesterol and PG values, had VLDL cholesterolffg ratios., had a much lower rate of correct estimation, as would be expected. The number of those with PG of.1-. gil was small (n = 1), but they were best estimated with TG/, but at a rate of only 1%, and markedly lower rates for all other factors. Above PG of. gil, the number of individuals was equally small (n = 1), and none of the factors correctly estimated any of the samples. The entire group of combined Table. Mean (± SD) Percent Deviation of Estimated LDL Cholesterol from Measured LDL Cholesterola Factor used In estimation TO, g/l n TG/ TG/. TG/ TG/. TG/ TO/i TG/ 17*b 7b 7b 7b b. ± ± 1*b b #{7}gb.1-1. 19 ± ±9 7±9 1.1-1. 11 #{7}*1 b ± ± 7± 9± 1.1-. 1 *b b ± ± 9± 11±.1-. 9-7±11 11*b ±1 7±11 11±11 1±11.1-. ±1 -±1 #{7}*b ±1 9± 1±1 ±1.1-. 9-17 ± 1 - ± *b ± * ± ± 1 ± 1 ±.1-. 1 - ± ± *) 1 ± 7 ± 9 ± ± 9 7 ± 7.1-1. -7 ± - ± 7 - ± ± 1 ± 9 ± ± 9 1.1-1. - ± - ± 9 7 ± 7* ± 9 7 ± 7 97 ± 9 17 ± 1 >1. -1 ± 17-799 ± 1 - ± -7 ± -1 ± 7 - ± 1 7 ± 1 a Within each TG range,* indicatesthe estimation factor withthe lowest percent error. b Estimation factors where the mean estimated LDL cholesterol and the meanmeasuredldl cholesterolconcentrationsvaried by.o g/l. CLINICAL CHEMISTRY, Vol., No. 1, 199

Table. Percent of Samples for Which LDL Cholesterol Was within 1% of Measured LDL Cholesterol, Grouped by Lipldemic Type Factor used In estimation Classif. TO, g/l. % Pop. 1/ 1/. TG/ TG/. TQ/ TO!? 1/ Normo. s. 7 * 7 Hyperchol. s. 19 91 9 9 7 HyperTGt. >. 1 7* Combinedcd s. >. 7* * 9 VLDLC/TG.. 1* c.d >.. aw,.j.ln each lipid category, * IndIcatesthe estimationfactorwith the greatest number of samples with less than 1% difference between measured and estimatedlolcholesterolconcentration.b Totalcholesterolconcentrations<. g/l. C Total cholesterol concentrations. g/ d1 concentrations between. and.g/l. hyperlipidemics with a VLDL cholesterolltg ratio. (potentially Type Ill individuals) represented only.% of our entire population, all of them coming from the HNRC population. The percentage of values classified correctly according to NCEP guidelines for estimated vs measured LDL cholesterol concentrations, utilizing all of the factors, are shown in Pables and, and Figure. Table provides the data for samples with PG s.o g/l and indicates that samples with the highest (1.9 g/l) and the lowest (<1. g/l) LDL cholesterol concentrations had the best rates of correct classification. This is due in part to misclassification errors in these ranges being unidirectional and in part to these categories containing more samples with concentrations that were farther from the decision levels of 1., 1., and 1.9 g/l. Samples with measured LDL cholesterol values in the middle ranges of 1. to 1.9 gil had the possibility of being misclassified bidirectionally. Pable B indicates that samples with measured LDL cholesterol concentrations differing by >.1 gil from a decision level were correctly classified -9% of the time, and that PG/. and PG/ were the two best estimation factors (9%). When mea- Table. Percent of Samples Properly Classified for LDL Cholesterol Content According to NCEP Guldellnes Classified by U/C Classified by estimation Estimation factors (% classified) g/l g/l n 1/ TG/. A) For TG. g/l <1. 9* 9 <1. 1.-1.9 1 1.-1.9 o o i.9 1/ 1/. 1/ 117 1/ 9 9 1 7 1 1 19 1.-1.9 <1. 1.-1.9 1.-1.9 i.9 119 1 7 1 7 7 7 9 1 1 1 7 1 1 1 1 1.-1.9 <1. 1.-1.9 1.-i.9 i.9 o o 1 1 7 79 1 7 1 * 77 1 19 7 1.9 Overall <1. 1.-1.9 1.-1.9 1.9 97 19 1 1 o 9 9 97 9 1 * 79 7 B) For measuredldl cholesterol<.1 or.1 gil froma decisionlevel <.1 11.1 7 9 9* Overall 9* 9 9 9 9 * 79 7 a WIthIneach NCEP category, * indicatestheestimationfactorwith the highestpercentofproperlyclassifiedsamples. In Tables and, the primary line In each groupingis in boldface, to simplifyunderstanding. U/C, ultracentrlfugation. CLINICAL CHEMISTRY, Vol., No. 1, 199 9

Classified by U/C Table. Percent of Samples Properly ClassIfiedfor LDL Cholesterol According to NCEP Guidelines, for TG >. g/la Classified by Estimation Estimation Factors (% classified) g/l g/l A) For TG,.1-. gil <1. <1. 1.-1.9 1.-1.9 1.9 n 1/ 1/. 1/ 1/. 1/ 117 1 9* 1 1 7 1 1 9 1 1.-1.9 <1. 1.-1.9 1.-1.9 1.9 1 1 1 1 1 * 1 1 1.-1.9 <1. 1.-1.9 1.-1.9 1.9 1 7 7 7 * 7 7 1.9 Overall <1. 1.-1.9 1.-1.9 i.9 7 17 1 7 9 7 1* 1* B) For TG >. <1. <1. 1.-1.9 1.-1.9 i.9 79 9* 9 1 9 7 1 11 9 1 1.-1.9 <1. 1.-1.9 1.-1.9 i.9 1.-1.9 <1. 1.-1.9 1.-1.9 i.9 1 1.9 Overall <1. 1.-1.9 1.-1.9 i.9 7 91 a Within each NCEPcategory, * Indicatesthe estimationfactorwith the highest percent of properlyclassifiedsamples. U/C, ultracentrifugation. 9 1 1 sured concentrations were within.1 g/l of a decision level, however, samples were correctly classified in only -9% of cases, with TG/ having the highestrateat 9%. Data for samples with PG of.1-. g/l (Table A), and for samples with TG >. gil (Table SB) show similar distribution patterns. In all cases, regardless of PG concentrations, samples with the lowest measured LDL cholesterols were best classified with a low estimating factor (i.e., TG/) and, conversely, those with the highest measured values were best classified with a high factor (i.e., TG/). Of course, using TG/ causes one to estimate a lower LDL cholesterol and, conversely, using TG/ causes one to estimate a higher LDL cholesterol value for any given set of variables. Therefore, these latter findings are not surprising. Overall, PG/. was the best factor for PG values in the range.1-. g/l, and PG/ was the best classification factor when PG values were >. g/l. The number of samples in both of these groups was quite small, however, and the data were skewed, particularly in the lattergroup.properclassification into NCEP categories should not supercede the methods for most nearlyaccurate estimation, and our data indicate that at PG values >. g/l none of these methods allows for accurate estimates. At PG <. g/l, the factor PG/ provides a reasonable assessment of LDL cholesterol. CLINICALCHEMISTRY, Vol., No. 1, 199

Cl) >. -J (1) -j. C/) LL z 1 NCEP LDL CHOLESTEROL CUTPOINTS Fig.. Effect of estimatedldlcholesterolconcentrationson correct NCEP classificationfor each estimationfactor(tgi, TGI., TGI, TGI., TGI, TG/7, TG/) Open bars representsamples with TG c. gil; filled bars represent samples with TG.1-. gil Eachgroupof barsrepresentsone category of riskas definedby NCEP Discussion LDL cholesterol concentration has been recommended by the NCEP Adult Preatment Panel as the determining factor in initiating dietary and drug treatment (9), so it is important to understand the validity and limitations of reported values for LDL cholesterol. There is no method currentlyavailable for measuring LDL cholesterol directly from plasma or serum; thus this analyte must be measured after its separation by ultracentrifugation. Ultracentrifugation methodology was used by the Lipid Research Clinics in the Prevalence and Primary Prevention studies (,,1) and is generally regarded as the standard reference method. The ultracentrifugation technique depends on the accuracy of centrifugation, HDL precipitation, and the cholesterol assay. Owing to the expense and lack of availability of ultracentrifugation, however, methods were developed to estimate LDL cholesterol concentration. These estimation techniques must rely on the accuracy of the cholesterol and TG assays, the HDL precipitation, and also on an additional mathematical factor that is used in an attempt to estimate the VLDL cholesterol concentration. There has been some controversy as to the most nearly accurate estimation factor to use, and attempts have been made to evaluate and refine the original estimation calculations proposed by Friedewald et al. to decrease potential estimation errors (1, 1, 17-19). However, the most frequently used estimation method in clinical laboratories continues to be that of Friedewald et al. The present study was designed to re-evaluate the Friedewald and other estimation factors with regard to their impact on NCEP guidelines, and to examine the level of potential error associated with these mathematical estimation factors. We have previously confirmed that the Friedewald formula of TG/ is an appropriate factor for estimating VLDL and LDL cholesterol in individuals with PG. g/l (). Additionally, PGI also gave approximations of LDL cholesterol for individuals with elevated PG, but with a large margin of error (1). In the present study, we have re-examined those issues, using both normolipidemic and dyslipidemic adults (1-1 y), in conjunction with the examination of other calculation factors and NCEP adult classification guidelines. In addition, we have used automated enzymatic lipid assays of the type used in most clinical chemistry laboratories, including unblanked TG measurements, which differ from chemical methods used in our previous studies and in the Lipid Research Clinic studies. Because the identical values for total cholesterol, TG, and HDL cholesterol were used for both measuring and estimating each sample, and because these assays were CDC standardized, errors due to inaccuracy, imprecision, and biological variability were not assessed in this study. In clinical practice, however, variability must be minimized as much as possible with regard to assay reliability, duration of fasting before sampling (,), and interlaboratory standardization (). When comparing the estimated values with the measured values, we examined the data from two perspectives. The first analysis assessed how closely each estimated LDL cholesterol values approximated the measured value. For this we used a 1% deviation from the measured concentration as a cutpoint. This 1% range represented an allowable LDL cholesterol fluctuation of approximately.1 to. g/l (mean =. ±. gil; range =. to.9 g/l). Not surprisingly, the ability to correctly estimate LDL cholesterol concentration varied inversely with the PG concentration in the plasma. As seen in Table 1 and Figure 1, the proportion of samples falling within the fairly broad 1% range decreased continuously with increased PG concentration, irrespective of the calculation factor used. The standard deviation of the variation increased from ±7% for PG. gil, to ±1-1% at TG concentrations of.1-. g/l (Table ). Above. g/l, the standard deviation increased to ±% to more than ± 1% error. When we subdivided individuals according to their type of dyslipidemia (Table ), we found, as expected, that those with low PG concentrations (i.e., normolipidemics and hypercholesterolemics) usually were correctly estimated. Those with increased TG concentrations (i.e., hypertriglyceridemics and combined hyperlipidemics), however, were much less often correctly estimated, and the subgroup of combined hyperlipidemics with an abnormally high VLDL cholesterolffg ratio even more so. Fortunately, individuals in the latter subgroup are encountered infrequently in the general population-.% in our study-and their treatment would not differ substantially from that used for other combined hyperlipidemics. Unfortunately, these individuals are impossible to distinguish from other combined hyperlipidemics without ultracentrifugation and apolipoprotein E phenotyping. Because the NCEP Adult Treatment Panel guidelines are now being used as a basis for identifying and determining treatment of dyslipidemic individuals, our second assessment involved an examination of the accuracy of classification of LDL cholesterol values, derived through estimation, into various NCEP categories. Because the data used by NCEP for LDL cholesterol cutpoints were derived primarily from samples analyzed after ultracentrifugation, we believed it important to examine how individuals would be classified if ultracentrifugation were not available. Interestingly, as shown in Tables and and Figure, most individuals were properly classified, even though the esti- CLINICALCHEMISTRY, Vol., No. 1, 199 1

mated LDL cholesterol concentrations were less accurate than one would wish. Those with values farthest from a decision level (i.e., very low or very high values) were most accurately classified. Individuals with values close to the decision points, where accuracy is most critical, were naturally at the highest risk of misclassification (Table B). Because of this, it is particularly important that these individuals have their values rechecked to minimize inappropriate counseling. Additionally, it is extremely important that patients adhere to 1- to 1-h fasts, because those with higher PG concentrations are in double jeopardy of being misclassified: they (a) are subject to a higher rate of inaccurate LDL cholesterol estimation due to the elevation in PG and (b) are generally more sensitive to changes in plasma PG from dietary fat intake, often requiring a longer time to return to baseline after a meal (,). Our data clearly indicate the need for a convenient direct assay for measuring LDL cholesterol. However, at the present time, utilization of the original formula of Friedewald et al., with the factor TG/, appears to provide a reasonable alternative for individuals with PG <. g/l. For those with PG >. g/l, TG/ is still the factor that gives results most closely matching the measured LDL cholesterol, but the potential error is so great for any given individual that we do not recommend its use. It should be noted that the prevalence of LDL cholesterol values 1. g/l in subjects with TO >. gil is reasonably low (%, Table ). We recommend that these individuals (.% of our entire study; 1.% of the normal Framingham population) and those suspected of having Type ifi hyperlipoproteinemia be referred to specialized laboratories with facffities available for ultracentrifugal analysis if the type of lipsprotein abnormality is to be properly assessed (9). We thank Dr. Mary M. Schaefer (HNRC) for carrying out the statistical analysis, and Albina Mariano and Babe Bravo for performing all of the lipid and lipoprotein determinations for the FHS samples. This work was supported by GrantHL and Subcontract RFP NHLBI HV - from the National Heart, Lung and Blood Institute, and Contract -K--1O from the U.S. Dept. of Agriculture Research Service. References 1. Keys A, ed. Coronary heart study in seven countries. Circulation 197;1(Suppl):I1-11.. Martin M,J, Hulley SB, Browner WS, Kuller LH, Wentworth D. Serum cholesterol, blood pressure, and mortality: implications from a cohort of 1, men. Lancet 19;ii:9-.. Castelli WP, Garrison RJ, Wilson PWF, Abbott RI), Kalousdian S, Kannel WB. Incidence of coronary heart disease and lipoprotein cholesterol levels. J Am Med Assoc 19;:-.. Lipid Research Clinics Program. The Lipid Research Clinics Coronary Primary Prevention Trial results: I. Reduction in the incidence of coronary heart disease. J Am Med Assoc 191:1-.. Lipid Research Clinics Program. The Lipid Research Clinics Coronary Primary Prevention Trial results: IL The relationship of reduction in incidence of coronary heart disease to cholesterol lowering. J Am Med Assoc 19;1:-7.. Brensike JF, Levy RI, Kelsey SF, et al. Effects of therapy with cholestyramine on progression of coronary arteriosclerosis: results of the NHLBI Type II Coronary Intervention Study. Circulation 19;9:-. Elo, Haapa K, trial with with dyslipidemia. Safety of treatment, changes in risk factors, and incidence of coronary heart disease. N Engl J Med 197;17:17-. 7. Frick MB, primary-prevention et al. Helsinki gemlibrozil in middle-aged Heart Study: men. Blankenhorn DH, Nessim SA, Johnson RL, Sanmarco ME, Azen SP, Cashin-Hemphill L. Beneficial effects of combined colestipol-niacin therapy on coronary atherosclerosis and coronary venous bypass grafts. J Am Med Assoc 197;:-. 9. Expert Panel: Report of the National Cholesterol Education Program expert panel on detection, evaluation, and treatment of high blood cholesterol in adults. Arch Intern Med 19;1:-9. 1. Havel RJ, Eder HA, Bragdon JH. The distribution and chemical composition of ultracentrifugally separated lipoproteins in human serum. J Clin Invest 19;:-. 11. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol without use of the preparative ultracentrifuge. in plasma, Cliii Chem 197;1:99-. 1. DeLongDM, DeLongER, Wood PD, Lippel K, Rilkind BM. A comparison of methods for the estimation of plasma low- and very low-density lipoprotein cholesterol: the Lipid Research Clinics Prevalence Study. J Am Med Asaoc 19;:7-7.. McNamara JR, Schaefer EJ. Automated enzymatic standardized lipid analyses for plasma and lipoprotein fractions. Clin Chim Acts 197;1:1-. 1. Warnick GR, Benderson J, Albers JJ. Dextran sulfate-mg precipitation procedure for quantitation of lipoprotein cholesterol. Clin Chem 19;:79-. high-density. 1. Kuba K, FrantzJr ID. Validity of the Friedewald equation for the estimation of plasma low-density lipoprotein cholesterol (LDL- C) at various concentrations of triglycerides [Abstract]. Clin Chem 197;:9. 1. The Lipid Research Clinics Program. Population Studies Data Book, vol. 1. The prevalence study. NIH publ. no. -17. Washington, DC: US Govt Printing Office, 19. 17. NiedbalaRS, Schray KJ, Foery R, Clement G. Estimation of low-density lipoprotein by the Friedewsid formula and by electrophoresis compared [Letter]. Cliii Chem 19;1:17.-. 1. Lippi U, Graziani ME, Manzato F, Schinella M. The Friedewald formuim statistical acrobatics? [Letter]. Clin Chem 19;:99-1. 19. Rao A, Parker All, El-Sheroni NA, Babelly MM. Calculation of low-density lipoprot.ein cholesterol with use of triglyceride/ cholesterol ratios in lipoproteins compared with other calculation methods. Clin Chem 19;:-.. Wilson PW, Abbott RI), Garrison RJ, Castelli WP. Estimation of very-low-density lipoprotein cholesterol from data on triglyceride concentration in plasma. Cliii Chem 1917:-1. 1. Wilson PWF, Zech LA, Gregg RE, et al. Estimation of VLDL cholesterol in hyperlipidemia.clin Chin Acts 19;11:-91.. CohnJS, McNamara JR, Schaefer EJ. Lipoprotein cholesterol concentrations in the plasma of human subjectsasmeasured in the fed and fasted states. Cliii Chem 19;:-9.. CohnJS, McNamara JR. Cohn SD, Ordovas JM, Schaefer EJ. Postprandial plasma lipoproteinchanges in human subjects of different ages. J Lipid Rae 19;:9-.. Current status of blood cholesterol measurement in clinical laboratories in the United States: a report from the Laboratory Standardization Panel of the NationalCholesterol Education Program. Clin Chem 19;:19-1. CLINICALCHEMISTRY,Vol., No. 1, 199