Traditional modifiable risk factors for cardiovascular disease (CVD) are smoking, high blood pressure, and unfavorable blood lipid concentrations. Models combining these factors predict CVD more accurately than models considering CVD risk factors in an isolated manner.1-3 Combined risk prediction models include the Framingham Risk Score or, from Europe, the SCORE (Systematic Coronary Risk Evaluation).1,2 One disadvantage of these assessments is that they require blood sampling for lipid measurements. This precludes the estimation of the 10-year risk of a CVD event, eg, from self-reports. In electronic health records, the lack of information on cholesterol was the most common reason why CVD risk could not be calculated.4 In contrast, body height and weight are available in virtually all health data sets. On the basis of the SCORE method and using a population sample from Switzerland, we aimed at comparing the traditional prediction model using total cholesterol with a version in which we replaced cholesterol with body mass index (BMI).1
Risk factor data stem from 17 791 men and women older than 16 years who participated in either of 2 CVD studies: the National Research Program 1A (NRP1A), a community health promotion initiative focused on CVD prevention, and the Swiss MONICA (Monitoring of Trends and Determinants in Cardiovascular Disease) population survey, an international project of the World Health Organization. We obtained mortality follow-up by anonymously linking the data from the CVD studies with the Swiss National Cohort (SNC), which encompasses all residents of Switzerland enumerated in the national 1990 or 2000 censuses as well as data from death and emigration registries until the end of 2008. Linkage success was 94% (NRP1A) and 97% (MONICA). The 95th percentile of follow-up was 31.2 years, during which 2170 men and 1761 women died (749 and 630 from CVD, respectively).5,6
Blood sampling and cholesterol measurement were described.5,6 Body mass index was calculated from measured (without shoes) height and weight (calculated as weight in kilograms divided by height in meters squared). We defined smoking as smoking 1 cigarette or more per day. Nonsmokers include former and never smokers. Systolic blood pressure was recorded as the mean of 2 measurements. Fatal CVD events were defined according to the Eighth Revision International Classification of Diseases codes 390 to 458 (until 1994) and International Statistical Classification of Diseases, 10th Revision codes I00 to I99.
Risk models were calculated with Weibull proportional hazards regression as previously described.1 To compare the prediction abilities of the cholesterol and BMI model, we calculated the mean cross-validated (leave-one-out) Brier score,7 which measures the mean squared difference between the risk score and the actual outcome. The lower the difference, the better the respective risk prediction model. The Brier score covers both calibration and sharpness of a prediction model.7
Compared with cholesterol (eFigure), the BMI model (Figure) showed higher risks at all ages and could better discriminate persons at high and low CVD risk. Moreover, the synergistic effects in combination with smoking and in particular with blood pressure were stronger than with cholesterol. Body mass index, but not cholesterol, was significantly associated with mortality. The prediction ability of BMI was better based on the lower Brier score (eTable 1). Because explanatory variables (age, sex, smoking, and blood pressure) other than BMI or cholesterol remained the same in the 2 models, the difference between the Brier scores was small. In a common model with cholesterol, BMI remained significant, while cholesterol did not (eTable 2). Thus, cholesterol did not contribute to the explanation of the association between risk factors and mortality when BMI was included in the same model.
Using BMI instead of cholesterol in CVD risk prediction models may provide more accurate estimates. Traditional models such as Framingham or SCORE include cholesterol or total to high-density lipoprotein cholesterol ratio but do not consider BMI in their equation.1,2 In line with our results, Green et al4 found that using BMI instead of cholesterol allowed at least equivalent CVD risk estimation based on electronic health records and that the use of BMI could reduce unnecessary laboratory testing. The fact that BMI renders blood sampling unnecessary leads to a substantial increase of population-based samples available for CVD risk estimation. The use of BMI may not only ease CVD risk assessment but could have further advantages. Compared with dyslipidemia screening, screening for obesity has a stronger scientific foundation and is unconditionally recommended.4 Furthermore, lifestyle changes (diet and physical activity) promoting weight loss or preventing weight gain may improve health more strongly than lipid-lowering treatment. In contrast, knowledge of cholesterol may not lead to behavioral changes, and there are also doubts concerning the effectiveness and safety of statin treatment for primary prevention of CVD.4,8
In conclusion, our results suggest that BMI may be a valuable alternative to cholesterol in CVD risk prediction models. This finding needs to be validated in other populations.
Correspondence: Dr Faeh, Institute of Social and Preventive Medicine, University of Zurich, Hirschengraben 84, 8001 Zurich, Switzerland (email@example.com)
Published Online: November 12, 2012. doi:10.1001/2013.jamainternmed.327
Author Contributions: All authors had full access to the data. Study concept and design: Faeh and Bopp. Acquisition of data: Braun and Bopp. Analysis and interpretation of data: Faeh and Braun. Drafting of the manuscript: Faeh and Braun. Critical revision of the manuscript for important intellectual content: Faeh and Bopp. Statistical analysis: Braun. Obtained funding: Faeh and Bopp. Study supervision: Faeh and Bopp.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported by grants 3347CO-108806, 33CS30-134273, and 32473B-125710 from the Swiss National Science Foundation.
Additional Contributions: The Swiss Federal Statistical Office for providing mortality and census data.
Conroy RM, Pyörälä K, Fitzgerald AP,
et al; SCORE Project Group. Estimation of ten-year risk of fatal cardiovascular disease in Europe: the SCORE project. Eur Heart J
. 2003;24(11):987-100312788299PubMedGoogle ScholarCrossref
D’Agostino RB Sr, Grundy S, Sullivan LM, Wilson P.CHD Risk Prediction Group. Validation of the Framingham coronary heart disease prediction scores: results of a multiple ethnic groups investigation. JAMA
. 2001;286(2):180-18711448281PubMedGoogle ScholarCrossref
Jackson R, Lawes CM, Bennett DA, Milne RJ, Rodgers A. Treatment with drugs to lower blood pressure and blood cholesterol based on an individual's absolute cardiovascular risk. Lancet
. 2005;365(9457):434-44115680460PubMedGoogle Scholar
Green BB, Anderson ML, Cook AJ,
et al. Using body mass index data in the electronic health record to calculate cardiovascular risk. Am J Prev Med
. 2012;42(4):342-34722424246PubMedGoogle ScholarCrossref
Bopp M, Braun J, Faeh D, Gutzwiller F.Swiss National Cohort Study Group. Establishing a follow-up of the Swiss MONICA participants (1984-1993): record linkage with census and mortality data. BMC Public Health
. 2010;10:56220858236PubMedGoogle ScholarCrossref
Bopp M, Braun J, Gutzwiller F, Faeh D.Swiss National Cohort Study Group. Health risk or resource? gradual and independent association between self-rated health and mortality persists over 30 years. PLoS One
. 2012;7(2):e3079522347405PubMedGoogle ScholarCrossref
Gneiting T, Raftery AE. Strictly proper scoring rules, prediction and estimation. J Am Stat Assoc
. 2007;102:359-378Google ScholarCrossref
Taylor F, Ward K, Moore TH,
et al. Statins for the primary prevention of cardiovascular disease. Cochrane Database Syst Rev
. 2011;(1):CD00481621249663PubMedGoogle Scholar