Discordance between regression-derived and ATP III table-derived Framingham 10-year risk: A P.A.T.H. Substudy

Purpose: Current U.S. guidelines for treatment of high cholesterol (ATP III) recommend LDL goals based on individual Framingham 10-year “hard” CHD Risk (FR). ATP III guidelines provide tables for estimating FR, but these tables only approximate the risk estimated by the regression-derived coefficients from the Framingham cohort. The concordance between these estimates of FR is not well known.

Methods: Medical records from 110 patients in a general Family Medicine practice were randomly selected for data abstraction in this cross-sectional analysis. For each patient, data were collected for gender, age, systolic blood pressure, presence/absence of antihypertensives, total cholesterol, HDL, and smoking status. Two FR scores were calculated per patient using the ATP III tables and a mathematical algorithm derived from the Framingham study. The difference and absolute difference in FR calculations for each patient were determined and a correlation analysis between the two FR calculations was performed.

Results: Using the ATP III tables, the mean and median FR were calculated as 13% and 11% (IQR, 6% to 20%), whereas the regression-derived mean and median FR were calculated as 13.4% and 11.6% (IQR, 5.3% to 20.1%). Although the overall mean difference between FR calculations was small (0.34%), we found a significant mean absolute difference between FR calculations of 2.5% (IQR, 0.5% to 3.6%) with a range of 0.02% to 32.3% (p<0.0001 for the test that the mean absolute difference equaled 0). Ten percent of patients had an absolute difference between FR calculations exceeding 5%. Additionally, 24.5% of patients were categorized differently (e.g. low/moderate/high risk) depending on the FR calculation used. The FR calculations were highly correlated (r=0.88; p<0.001).

Conclusion: ATP III table calculations of FR differ significantly from the regression-derived FR in a significant proportion of patients. Additional studies will determine the impact of this discordance on prescribing patterns for cholesterol medications.