In this model, some adjacent risk factor categories were combined to avoid cells with limited numbers of events and/or unpredictive trends. To compare prediction of these three risk models, we examined different statistical measures. To assess discrimination, we used Harrell��s C-index, an adaptation of the C-statistic an adaptation of the C-statistic or area under the ROC curve for use with survival data. As the model validation for Health ABC functions was performed on the same dataset used for estimating the Cox model and the sample included too few events for splitsample validation, we calculated an optimism-corrected C-index using bootstrap resampling with 1000 replications. To assess model calibration, we used Parzen��s adaptation of the Hosmer-Lemeshow test to the Cox model. In exploratory analysis, we sought to determine whether alternative sets of predictors would improve risk prediction. To evaluate the utility of adding to the FRS different lifestyle and simple laboratory variables, we initially considered predictor variables with p,0.20 in unadjusted Cox models for CHD events in Health ABC data. We then used three model selection procedures: a backward selection with a retention criterion of p,0.10, and two forward stepwise selection procedures minimizing the Akaike Information Criterion and the Bayesian Information Criterion, respectively. We used a variety of model selection procedures when considering the addition of routinely available measures not included in the Framingham risk factor set to the Health ABC function. The procedures based on p-values and the AIC lead to very similar final models ; in contrast, the BIC, which strongly Folinic acid calcium salt pentahydrate penalizes the complexity of the model, lead to the omission of a larger number of risk factors. All final models mainly retained traditional risk factors included in the FRS. The additions of lifestyle variables, waist circumference, and creatinine did not improve risk prediction in terms of discrimination or model fit beyond using the traditional risk factors from the FRS. Selection procedures stratified by gender yielded similar results. In this population-based study of older adults, the FRS poorly discriminated between persons who experienced a CHD event and those who did not and underestimated the absolute CHD risk by 51% in women and 8% in men. Nevertheless, traditional risk factors remained the best predictors of CHD events. Physical activity, alcohol consumption, waist Tulathromycin B circumference and creatinine did not improve risk prediction beyond traditional risk factors of the FRS. Recalibration of the FRS improved the accuracy of absolute risk estimation, particularly for women. For both genders, the Health ABC function significantly improved estimation of absolute risk, with a discrimation similar to the FRS. Neither refitting equations nor including other routinely available measurements in risk equations provided substantial benefits in terms of discriminating between high and low-risk.