(2006) 368–371www.elsevier.com/locate/ypmed

Preventive Medicine 43

National Cholesterol Education Program risk assessment andpotential for risk misclassification

Stephen D. Persell a,⁎, Donald M. Lloyd-Jones b,c, David W. Baker a

a Division of General Internal Medicine, Northwestern University, 676 N. St. Clair Street, Suite 200, Chicago, IL 60611-2927, USAb Division of Cardiology, Department of Medicine, Feinberg School of Medicine, Northwestern University, Chicago, IL 60611-2927, USA

c Department of Preventive Medicine, Feinberg School of Medicine, Northwestern University, Chicago, IL 60611-2927, USA

Available online 14 August 2006

Abstract

Background. The National Cholesterol Education Program Adult Treatment Panel report from 2001 (ATP III) recommends clinicians calculate10-year coronary risk using multivariable methods only for adults with 2 or more risk factors. We aimed to determine who would be falselyclassified as low risk using this approach.

Methods. We studied 4097 adults aged 20 to 79 years without diagnosed cardiovascular disease or diabetes from the National Health andNutrition Examination Survey from 1999 to 2002. We determined the proportion with fewer than 2 risk factors who nonetheless had estimated 10-year risk of cardiac death or myocardial infarction ≥10% using multivariable methods.

Results. Among persons with fewer than 2 risk factors, 5.3% (95% confidence interval 4.7 to 6.1%), had a 10-year risk ≥10% using theFramingham Risk Score and would be misclassified using the risk factor counting method (this corresponds to approximately 5,640,000 U.S.adults). Compared to individuals whose classification was unchanged, those misclassified as low risk were older (P<0.001) and more likely male(85.5% vs. 41.2%, P<0.001).

Conclusions. Relying on the ATP III risk factor counting method rather than determining risk using multivariable methods in all patientsresulted in misclassifiying as low risk over 5 million adults with at least moderately high risk of coronary heart disease, most of whom are middle-aged and older men.© 2006 Elsevier Inc. All rights reserved.

Keywords: Hypercholesterolemia; Cardiovascular diseases; Practice guidelines; Risk; Risk factors

Introduction

The National Cholesterol Education Program (NCEP) AdultTreatment Panel report from 2001 (ATP III) recommends riskassessment using risk factor counting for adults withoutestablished cardiovascular disease (CVD) or diabetes mellitus.Risk assessment is used to help clinicians decide on low-densitylipoprotein cholesterol (LDL-C) treatment goals for the primaryprevention of cardiovascular disease. Clinicians are advised toestimate 10-year risk of fatal coronary heart disease (CHD) ornonfatal myocardial infarction using tables derived from theFramingham Heart Study only for adults with 2 or more riskfactors. Adults with fewer than 2 risk factors, and those with 2risk factors and elevated high-density lipoprotein cholesterol

⁎ Corresponding author. Fax: +1 312 695 0951.E-mail address: spersell@nmff.org (S.D. Persell).

0091-7435/$ - see front matter © 2006 Elsevier Inc. All rights reserved.doi:10.1016/j.ypmed.2006.06.017

(HDL-C) are assumed be at low risk, and risk estimation forthese groups is not routinely recommended (Expert Panel onDetection, Evaluation, And Treatment of High Blood Choles-terol In Adults (Adult Treatment Panel III), 2001, 2002).Cholesterol lowering with medication in the low-risk group isonly recommended for individuals with very high LDL-Clevels.

The ATP III report stated that risk calculation for the low-riskgroup was not necessary because the 10-year risk rarely reachesa level high enough to warrant intensive cholesterol treatment.The panel estimated that 2.6% of the group with fewer than 2risk factors would have an estimated 10-year coronary risk of atleast 10% using the multivariable method derived from theFramingham study and therefore be misclassified by risk factorcounting (Expert Panel on Detection, Evaluation, And Treat-ment of High Blood Cholesterol In Adults (Adult TreatmentPanel III), 2002). Due to the aging of the United States

369S.D. Persell et al. / Preventive Medicine 43 (2006) 368–371

population and the high prevalence of borderline abnormalcardiovascular risk factors, this approach may falsely classifymore individuals as low risk than was previously thought.

We sought to determine the proportion of the U.S. populationdesignated as low risk using the ATP III risk factor countingmethod who have an estimated 10-year risk of fatal CHD ornonfatal myocardial infarction of ≥10%, and to describe thecharacteristics of individuals most likely to be mistakenlyclassified as low risk using the risk factor counting method.

Methods

We used data from the National Health and Nutrition Examination Surveyconducted in1999 to 2002 obtained from a nationally representative sample ofthe noninstitutionalized U.S. population. Participants completed a homeinterview, physical examination and laboratory testing. The survey used amultistage sampling design and over-sampled non-Hispanic blacks, MexicanAmericans, persons over 60 years old, and low-income individuals. Theinstitutional review board of the National Center for Health Statistics approvedthe survey procedures and participants provided informed consent (NationalCenter for Health Statistics, 2005). We included nonpregnant participants whowere 20 to 79 years old, did not report physician diagnosed coronary arterydisease, stroke, or diabetes mellitus, and were not currently using cholesterollowering medication. We excluded persons who did not have cholesterol orblood pressure measurements, or had missing smoking information. Weclassified subjects into 4 cardiovascular risk groups according to the methodsin the NCEPATP III report using risk factor counting and estimation of 10-yearrisk of fatal CHD or nonfatal myocardial infarction: (1) low risk (0–1 riskfactor), (2) moderate risk (≥2 risk factors and <10% 10-year risk), (3)moderately high risk (≥2 risk factors and 10–20% 10-year risk), and (4) highrisk (≥2 risk factors and >20% 10-year risk) (Expert Panel on Detection,Evaluation, And Treatment of High Blood Cholesterol In Adults (AdultTreatment Panel III), 2001). The following risk factors were summed: male sex≥45 years old, female sex ≥55 years old, a first-degree relative with prematureCHD (only data for parents and siblings with myocardial infarction or anginabefore age 50 years was available), current cigarette smoking, hypertension(systolic ≥140 or diastolic ≥90 mm Hg or taking anti-hypertensivemedication), or serum high-density lipoprotein cholesterol (HDL-C) <40 mg/dl. The risk factor sum was reduced by one if the HDL-C was≥60 mg/dl. Thosewith a sum less than 2 were assigned to the “low-risk” group and constitute thestudy cohort (n=4097).

Ten-year risk of fatal CHD or nonfatal myocardial infarction was determinedusing a multivariable risk assessment tool derived from Framingham HeartStudy and published in the ATP III report for clinical use (Expert Panel onDetection, Evaluation, And Treatment of High Blood Cholesterol In Adults(Adult Treatment Panel III), 2001). Individuals with a risk factor sum of <2 withan estimated 10-year risk of ≥10% were considered to be misclassified. Wedetermined the proportion of adults from the low-risk group who weremisclassified separately for men and women in selected age strata.

We used SAS release 9.1 (SAS Institute Inc., Cary, NC) and SUDAANrelease 9.0.0 (Research Triangle Institute, Research Triangle Park, NC) withappropriate sampling weights (which account for the probability of subjectselection and nonresponse) to account for the survey design and to providenationally representative proportions and accurate confidence intervals. We usedthe χ2 test to compare categorical variables between participants who werecorrectly classified as low risk, and those who were misclassified. Tests ofsignificance are 2-sided with a P value set at <0.05.

Results

Using risk factor counting (the method recommended by theATP III), 76.1% of adults aged 20 to 79 without diagnosedcoronary artery disease, stroke or diabetes were classified as lowrisk, 13.5% were moderate risk, 10.2 were moderately high risk,

and 3.8% were high risk. Among the “low-risk” individuals,5.3% (95% confidence interval 4.7 to 6.1%) had an estimated10-year risk ≥10% risk using the Framingham method. Thisproportion corresponds to 5,640,000 adults in the United Statespopulation (Table 1).

Compared with correctly classified low-risk individuals,persons misclassified as low risk by the risk factor countingmethod were older, more often male, had higher blood pressure,higher total and non-HDL cholesterol and were more oftensmokers (Table 1). Nearly all (98%) correctly classified adultswere under age 65 while 71.3% of misclassified persons wereunder age 65. Eighty-six percent had total cholesterol levels ofat least 200 mg/dl and 65% had non-HDL cholesterol levels ofat least 160 mg/dl. While 62.8% of properly classified personshad ideal systolic blood pressure, 76.8% of misclassified adultshad systolic blood pressure ≥120 mm Hg.

The estimated 10-year risk of fatal CHD or nonfatalmyocardial infarction was low for properly classified low-riskindividuals with 73.2% having a 10-year predicted risk less than2% and 94.1% having a risk ≤6%. Most misclassifiedindividuals (76.7%) had a 10-year predicted risk of 10 to12%, and 6.8% had a risk of at least 20%. Thirty-one percent ofmisclassified subjects had 2 risk factors and an HDL cholesterollevel ≥60 mg/dl (Table 1).

The distribution of misclassified adults by age and sex stratais depicted in the Fig. 1. Almost half of misclassified adultswere men aged 40 to 64 years.

Discussion

Although assumed to be at low risk by the NCEPATP III riskfactor counting method, 5.3% of this population have at least a10% predicted 10-year risk of CHD death or nonfatalmyocardial infarction. Risk classification using risk factorcounting misclassified significant numbers of men in all agegroups and women over 65 years of age. Most misclassifiedpersons have moderately elevated cholesterol levels.

Risk estimation is used to determine the goals ofcholesterol-lowering treatment for the primary prevention ofcardiovascular disease in adults. Risk factor counting isrecommended by the ATP III as the initial step in riskestimation for adults without CVD or diabetes mellitus,although the full NCEP report does acknowledge theacceptability of using multivariable risk estimation as theinitial step in risk assessment (Expert Panel on Detection,Evaluation, And Treatment of High Blood Cholesterol InAdults (Adult Treatment Panel III), 2002). One advantage ofthe risk factor counting approach is its simplicity. Evenstraightforward multivariable methods such as the use ofinternet-based risk calculators, (National Heart Lung andBlood Institute, 2006) personal digital assistants, (Davidson,2002) or manual calculation using tables (Expert Panel onDetection, Evaluation, And Treatment of High Blood Choles-terol In Adults (Adult Treatment Panel III), 2001) may not beacceptable to busy clinicians. As electronic health recordsbecome more sophisticated, cardiovascular risk estimationcould be performed automatically, thereby relieving clinicians

Table 1Adults age 18 to 79 years from the National Health and Nutrition ExaminationSurvey, 1999 to 2002, classified as low risk by the Adult Treatment Panel IIIguideline a

Characteristic Correctlyclassified

Misclassified P

N 3800 297Size of correspondingU.S. population

99,990,000 5,640,000

% (95% confidence interval) 94.7 (93.9–95.4) 5.3 (4.7–6.1)

% %

Age, years <0.001<40 81.9 21.940–64 16.0 49.4≥65 2.0 28.7

Male sex 41.2 85.5 <0.001Race/Ethnicity <0.001White, non-Hispanic 70.7 79.8Black, non-Hispanic 10.1 7.6Mexican American 8.6 3.9Other 10.6 8.7

Systolic blood pressure,mm Hg

<0.001

<120 62.8 23.2120–139 31.7 58.4≥140 5.5 18.4

Medication for hypertension 5.1 15.7 <0.001Total cholesterol, mg/dl <0.001<200 55.5 17.6200–239 31.2 42.1240–279 11.0 31.0≥280 2.4 9.3

High-density lipoproteincholesterol, mg/dl

<0.001

<40 10.9 040–59 56.7 64.5≥60 32.4 35.5

Non-HDL cholesterol, mg/dl <0.001<130 39.4 14.3130–159 29.3 20.3160–189 19.7 30.3≥190 11.6 35.2

Current cigarette smoker 18.0 26.1 0.03Parent or sibling withpremature coronaryheart disease

1.2 0 <0.001

Qualifying ATP III risk factors <0.001None 49.6 0Age only 10.5 51.1Hypertension only 6.6 0Smoking only 16.5 17.5HDL <40 mg/dl only 10.9 0Family history only 1.1 0Two risk factors andHDL-C ≥60 mg/dl

5.0 31.4

Framingham risk score<1 51.0 01 22.2 02–5 20.9 06–9 5.9 010 0 39.611–12 0 37.113–19 0 16.5≥20 0 6.8

a Percentages are weighted to account for the survey design.

370 S.D. Persell et al. / Preventive Medicine 43 (2006) 368–371

of the burden of either entering data into a separate riskcalculating program or doing it by hand. In the meantime, ourdata suggest that physicians should perform multivariable riskestimation for all men and for women over 65 years of age,regardless of their burden of risk factors (for example a healthy65-year-old man with treated hypertension and a totalcholesterol level of 210 mg/dl, HDL of 65 mg/dl and systolicblood pressure of 133 mm Hg has an estimated 12% 10-yearrisk of CHD death or nonfatal myocardial infarction but wouldbe considered to be in the low-risk group using the risk factorcounting method).

By failing to detect a group of individuals with a 10–12%10-year coronary risk, the risk factor counting method reducesnumber of persons who are likely to need drug therapy toreach their LDL-C goals and therefore decreases the popula-tion-wide financial burden of treating hypercholesterolemia.Since this misclassified group has a moderate short-term riskof CHD, treating them with drug therapy may be less cost-effective than for higher risk groups (Jacobson et al., 1998).This rationale is insufficient to justify withholding considera-tion of lipid-lowering therapy for the misclassified group.However, data from randomized trials indicate that personswith event rates similar to those anticipated for themisclassified portion of the population derive considerablebenefits from cholesterol lowering therapy. For example, in arandomized primary prevention trial of lovastatin, amongindividuals where the placebo group's rate of CHD death,nonfatal myocardial infarction or unstable angina was 10.9 per1000 patient-years (similar to a 10.9% 10-year risk), and therewas an absolute reduction in these events of 4.1 per 1000patient-years. Coronary revascularizations were reduced by 3.1per 1000 patient-years (Downs et al., 1998). Furthermore, thefinancial burden associated with statin use should decrease aspatents for these drugs expire and their cost declines. Anotheroption for adults with intermediate cardiovascular risk is toundergo additional testing to help with risk estimation such asmeasurement of coronary artery calcium or other noninvasivemeasures of subclinical atherosclerosis (Greenland et al.,2004). But, if risk is assumed to be low based on risk factorcounting, it seems unlikely that this option will be consideredby clinicians.

Fig. 1. Estimation of number of U.S. adults misclassified as low risk by age andsex strata, National Health and Nutrition Examination Survey, 1999 to 2002.

371S.D. Persell et al. / Preventive Medicine 43 (2006) 368–371

Some misclassified individuals with very high LDL choles-terol would be candidates for cholesterol lowering drug treatmentbased on NCEP guidelines regardless of risk factor counting or10-year risk estimates. However, the majority who were mis-classified did not have extreme cholesterol elevations—60% hadtotal cholesterol levels below 240 mg/dl and 91% were below280 mg/dl. Furthermore, by classifying these persons as lowrisk, the target LDL cholesterol level would only be <160 mg/dl(Expert Panel on Detection, Evaluation, And Treatment of HighBlood Cholesterol In Adults (Adult Treatment Panel III), 2001)and some other appropriate risk reducing strategies, such asaspirin, may also not be adopted.

Our results should be interpreted with several limitationsin mind. We may have overestimated the number ofmisclassified individuals because we did not have completeinformation about family history of premature CHD. Usingself-reported family history of premature CHD for riskassessment is also limited by its inaccuracy (Murabito et al.,2004). On the other hand, by omitting from considerationadults aged 80 years and above with unfavorable cholesterolbut no other risk factors, it is likely that we underestimatedthe number of misclassified moderately high-risk adults inthe population.

In determining treatment thresholds for cholesterol in theprimary prevention of CVD, risk factor counting alone does notappear sufficient to identify all persons who may benefit fromcholesterol-lowering therapy. The NCEP ATP should advisephysicians to use multivariable risk assessment to estimatecoronary risk for all men over 40 and women over 65 years ofage in order to prevent the misclassification and under-treatmentof some patients. Employing computer information systems to

perform automated calculations would facilitate the process ofrisk assessment for all patients.

References

Davidson, M.H., 2002. Strategies to improve Adult Treatment Panel IIIguideline adherence and patient compliance. Am. J. Cardiol. 89, 8C–20C.

Downs, J.R., Clearfield, M., Weis, S., et al., 1998. Primary prevention of acutecoronary events with lovastatin in men and women with average cholesterollevels: results of AFCAPS/TexCAPS Air Force/Texas Coronary Athero-sclerosis Prevention Study. JAMA 279, 1615–1622.

Expert Panel on Detection, Evaluation, And Treatment of High BloodCholesterol In Adults (Adult Treatment Panel III), 2001. ExecutiveSummary of The Third Report of The National Cholesterol EducationProgram (NCEP). JAMA 285, 2486–2497.

Expert Panel on Detection, Evaluation, And Treatment of High BloodCholesterol In Adults (Adult Treatment Panel III), 2002. Third Report ofthe National Cholesterol Education Program (NCEP) Expert Panel onDetection, Evaluation, and Treatment of High Blood Cholesterol in Adults(Adult Treatment Panel III) final report. Circulation 106, 3143–3421.

Greenland, P., LaBree, L., Azen, S.P., Doherty, T.M., Detrano, R.C., 2004.Coronary artery calcium score combined with Framingham Score for riskprediction in asymptomatic individuals. JAMA 29, 210–215.

Jacobson, T.A., Schein, J.R., Williamson, A., Ballantyne, C.M., 1998.Maximizing the cost-effectiveness of lipid-lowering therapy. Arch. Intern.Med. 158, 1977–1989.

Murabito, J.M., Nam, B.H., D'Agostino Sr., R.B., Lloyd-Jones, D.M.,O'Donnell, C.J., Wilson, P.W., 2004. Accuracy of offspring reports ofparental cardiovascular disease history: the Framingham Offspring Study.Ann. Intern. Med. 140, 434–440.

National Center for Health Statistics, 2005. National Health and NutritionExamination Survey, viewed 1/26/05. http://www.cdc.gov/nchs/nhanes.htm.

National Heart Lung and Blood Institute, 2006. Risk Assessment Tool forEstimating 10-year Risk of Developing Hard CHD (Myocardial Infarctionand Coronary Death), viewed 4/25/06. http://hin.nhlbi.nih.gov/atpiii/calculator.asp?usertype=prof.