Nonlaboratory-based (non-LB) algorithms have been developed to facilitate absolute cardiovascular risk assessment in resource-constrained settings. The non-LB Framingham algorithm, which substitute BMI for lipids in laboratory-based Framingham, exhibits best performance among non-LB algorithms. However, its external validity has not been evaluated.Aim
To examine the validity of non-LB Framingham algorithm in Atherosclerosis Risk in Communities dataset, and contrast performance with the laboratory-based Framingham algorithm.Methods
We developed Cox regression models including non-LB and laboratory-based Framingham covariates in Atherosclerosis Risk in Communities dataset. Discrimination was assessed via C-statistic, calibration via goodness-of-fit, and marginal discrimination value of BMI vis-à-vis lipids vis-à-vis waist–hip ratio via net reclassification improvement (NRI). Both models were compared via area under receiver operating characteristic.Results
Among 11 601 participants (mean age 54 years, 55% women, 23% black), non-LB vs. laboratory-based Framingham performed as follows: C-statistic 0.75 vs. 0.76 among women and 0.67 vs. 0.68 among men; goodness-of-fit 14.2 vs. 10.5 among women and 25.8 vs. 21.8 among men. Overall area under receiver operating characteristic was 0.706 vs. 0.710, respectively, with no racial differences in discrimination or calibration. BMI and total cholesterol had no impact on NRI. Incremental predictive value of HDL was comparable with waist–hip ratio (category-less NRI = 0.34 vs. 0.31; categorical NRI7.5 = 0.06 vs. 0.05, P < 0.01).Conclusion
These results demonstrate the validity and limitations of the non-LB Framingham algorithm in a biracial cohort. Substituting BMI with a central adiposity metric such as waist–hip ratio or waist circumference could make the algorithm better or at par with the laboratory-based Framingham algorithm.