10 Year CVD Risk Peter W. F. Wilson, M.D. Emory University, Atlanta, GA
Disclosures Grants: Merck Consultant: Merck Janssen XZK Glaxo-Smith-Kline Astra-Zeneca
Risk Prediction Key Issues Risk Factors Specimen collection & storage (fluids, cells, other Variability (laboratory, biological) Standardized measurements Gradient of effects (linear, extremes, logarithmic) Clinically important Can generally intervene on factor Prefer low correlations between factors Special considerations for factors that might replace (e.g. apob for LDL-C) Avoidance of oversaturation of model Multiple measures of risk factors over time
CHD Prediction Score Sheet for Men--Cholesterol Categories Years Points 30-34 -1 35-39 0 40-44 1 45-49 2 50-54 3 55-59 4 60-64 5 65-69 6 70-74 7 Cholesterol mg/dl Points < 160-3 160-199 0 200-239 1 240-279 2 280+ 3 HDL-C mg/dl Points < 35 2 35-44 1 45-49 0 50-59 0 60+ -2 BP Dia BPDia PDia B BPDia BPDia BP Sys <80 80-8485-89 90-99 100+ <120 120-129 Optimal 130-139 Normal 140-159 High Normal 160+ Stage I Hypertension Stage II-IV Hypertension 0 pts Blood Pressure 0pt 1 pts 2 pts When systolic and diastolic pressures fall into different categories, use the higher category to classify the individual. 3 pts Adding up the points Diabetes Points No 0 Yes 2 Smoker Points No 0 Yes 2 Point Total 10 Year CHD Risk <= -1 2% 0 3% 1 3% 2 4% 3 5% 4 7% 5 8% 6 10% 7 13% 8 16% 9 20% 10 25% 11 31% 12 _ 37% 13 45% 14 >=53% Relative Risk Very Low Low Moderate High Very High Age (years) A n NHLBI Project Scoresheet developed by Peter W. F. Wilson, Ralph B. D'Agostino, Daniel Levy, Albert Belanger, Halit Silbershatz, and William B. Kannel Average Ideal* 10 Yr CHD10 Yr CHD Risk Risk 30-34 3% 2% 35-39 5% 3% 40-44 7% 3% 45-49 11% 4% 50-54 14% 5% 55-59 16% 6% 60-64 21% 8% 65-69 28% 10% 70-74 23% 13% Low risk person: BP < 120/80, HDL-C mg/dl(45 men, 55 women), LDL-C 100-129 mg/dl, non-smoker, no diabetes
CHD Prediction with Risk Factor Algorithms Risk Wilson ATP III D Agostino Assmann Euro-SCOR Factor 1998 2001 2001 2002 2003 Source Framingham Framingham Framingham PROCAM Europe Age interval 5 years 5 years 5 years 5 years 5 years Sex Yes Yes Yes Men Yes BP levels JNC-VI BP sys BP sys BP sys BP sys BP Rx No Yes No No No Cholesterol Yes Yes Yes No Yes HDL-C Yes Yes Yes Yes No LDL-C optional No No Yes No Cigarettes Yes Yes Yes Yes Yes Diabetes Yes No Yes Yes Yes ECG-LVH No No No MI Hx No Event Total CHD Hard CHD Hard CHD Hard CHD CHD Dth
Receiver Operating Characteristic Curves and Disease Prediction 1 Sensitivity (True Positives) 0.8 0.6 0.4 0.2 Better test Good test Chance Line 0 0 0.2 0.4 0.6 0.8 1 1-Specificity (False Positives)
1 ROC Curve for CHD Prediction Framingham Men 0.8 Sensitivity 0.6 0.4 Categories Continuous Scores Series2 0.2 0 0 0.2 0.4 0.6 0.8 1 False Positive Wilson Circulation 1998
What Counts? Absolute risk counts What will successfully reduce the absolute risk
Estimated 10 Hard CHD Risk Framingham Offspring and Cohort Men 100% Percent 80% 60% 40% 20% 10 Year Risk >20% 10-20% 6-10% <6% 0% 30-39 40-49 50-59 60-69 70-79 Age (years) Pasternak JACC 2003; 41: 1863
Estimated 10 Hard CHD Risk Framingham Offspring and Cohort Women 100% Percent 80% 60% 40% 20% 10 Year Risk >20% 10-20% 6-10% <6% 0% 30-39 40-49 50-59 60-69 70-79 Age (years) Pasternak JACC 2003; 41: 1863
Performance Measures for Risk Estimation Relative Risk Are the coefficients (Log RR) and study s optimal models the same within random fluctuations? Prediction of Outcomes: Discrimination Ability of the model to distinguish events from nonevents. (C-statistic used as a measure) Prediction of Outcomes: Calibration Closeness of predicted probability to observed (Adjusted Hosmer-Lemeshow Chi-Square <20 and calibration bar plot as measures) Serial Testing for Risk Estimation: Reclassification Bayesian method to test newer markers. Prior probability estimated with multivariable model. (Effects assessed with relative risk and pop risk)
Relative Risk for CHD Honolulu and Framingham Men 10 Year Follow up Optimal Blood Pressure Normal High Nl Stage 1 Stage 2+ Honolulu Framingham Other Factors Diabetes Smoking 0 0.5 1 1.5 2 2.5 3 D Agostino JAMA 2001; 286: 180 Relative Risk
Event free survival 1 0.98 0.96 0.94 0.92 0.9 Framingham and Honolulu 10 Year CHD Experience 0 1 2 3 4 5 6 7 8 9 10 Time (years) Honolulu Framingham D Agostino JAMA 2001; 286: 180
Honolulu Heart Study Hard CHD Prediction with Framingham Equations 5 Year Risk 20% 15% 10% 5% (no adjustment) Framingham Estimate Observed in Honolulu 0% 1 2 3 4 5 6 7 8 9 10 Decile of Risk D Agostino JAMA 2001; 286: 180
Honolulu Heart Study Hard CHD Prediction with Framingham Equations (adjustment for rates and risk factor levels) 5 Year Risk 20% 15% 10% 5% Framingham Estimate Observed in Honolulu 0% 1 2 3 4 5 6 7 8 9 10 D Agostino JAMA 2001; 286: 180 Decile of Risk
Probability Estimation of 10 Year Hard CHD Risk in Chinese Cohort Using Framingham CHD Functions 0.20 0.15 0.10 0.05 Men Predicted Actual Men--uncalibarated Probability 0.05 0.04 0.03 0.02 0.01 Predicted Actual Men--calibrated Men 0.00 1 2 3 4 5 6 7 8 9 10 0.00 1 2 3 4 5 6 7 8 9 10 Decile of predicted risk from Framingham functions Liu J JAMA June 2, 2004
Serial Testing and Risk of Disease Posterior Probability (%) 40 30 20 10 Test positive Test negative 0 0 10 20 30 Initial Probability (%) 40
Serial Testing and Risk of Disease Posterior Probability (%) 40 30 20 10 Test positive Test negative 0 0 10 20 30 Initial Probability (%) 40
Risk Reclassification and Prediction of CVD in US Adults Assessment Net Reclassification* C-Reactive Protein 5% to 15% Carotid IMT 10% to 15% Coronary Artery Calcification 25% * Basic models with age, sex, cholesterol, HDL-C, blood pressure, diabetes, smoking
CVD Prediction Special Populations Persons with limited life expectancy Adults with chronic diseases Diabetes Mellitus HIV Chronic Inflammation (SLE, RA etc) Chronic Kidney Disease Extremes of risk factors Familial Hypercholesterolemia
Cardiovascular Mortality in the General Population (NCHS) and in Kidney Failure Treated by Dialysis or Transplant (USRDS) Sarnak Circulation 2003; 108:2154
Are Risk Factor Scoring Systems For CVD Useful? Predictive Accuracy Discrimination is high. Calibration improves accuracy. Generalizability Approaches have been applied to other populations Variety of risk factor scores are available Cost Risk scoring is generally inexpensive Methods Can Assess New Factors Risk scoring is a dynamic field. Newer methods are being developed Safety Low risk (history, physical exam, blood tests) No radiation exposure Health Provider Relationship Score provides composite measure to help guide care
CVD Prediction Unresolved Issues and Future Directions Risk Estimation without a Clinical Visit Personal information and healthy behaviors to estimate CVD risk Inclusion of Genetic Marker Data Family History of CVD Self report vs validated report Parental information Sibling information CVD Risk Equivalents Diabetes Mellitus Chronic Kidney Disease Risk Factor Extremes Role of Excess Adiposity Prediction of Recurrent Cardiovascular Disease Replacing Risk Factors in Prediction Models
Characteristics of the Eight Americas in the United States Group Description Population (Millions) Income per capita (US Dollars) Finishing High School (per cent) 1 Asian 10.4 $21,566 80% 2 Low income rural white 3.6 $17,758 83% 3 Middle America not Black 214.0 $24,640 84% 4 Low income White Appalachia 16.6 $16,390 72% 5 Western Native American 1.0 $10,029 69% 6 Black Middle America 23.4 $15,412 75% 7 Southern low income rural Black 5.8 $10,463 71% 8 High risk urban Black 7.5 $14,800 72% Murray PLoS Med 2006; 3: e260
Life Expectancy at Birth in the Eight Americas within the United States Men (Asian) (Middle America) Black Middle America Murray PLoS Med 2006; 3: e260
Life Expectancy at Birth in the Eight Americas within the United States Women (Asian) (Middle America) Black Middle America Murray PLoS Med 2006; 3: e260
Summary Prediction of an Initial CVD Event Characteristic 5 to 15 Year Risk Age Group (yr) 40-75 Risk Factors Imaging (carotid IMT, coronary calcium) Traditional (+/- CRP) Potentially useful Exceptions Familial Hypercholesterolemia Severe Chronic Kidney Disease Limitations Data mostly for whites