The demographic characteristics of the study sample are as shown in . Half (50.2%) of the cohort was female. There were 34.3% Caucasians, 19.7% Chinese, 20.8% African Americans and 25.1% Hispanics. A total of 182 (6.0%) subjects had an adjudicated CVD event over a maximum of five years [Incident rate of adjudicated CVD event in the entire MESA study (n=6814) was 2.6% over the five years of follow-up]. Traditional risk factors were higher in cases than the sub cohort ().
Demographic characteristics [mean ± SD or n (%)] of the subcohort (n=2843) and incident cases (n=182), MESA.
Subjects with sex-specific FMD responses higher than the median had fewer events than those with sex-specific FMD responses lower or equal to the median during 5 years of follow-up (log rank p-value <0.0001, ). In univariate probability-weighted Cox proportional hazard analysis, FMD/unit SD was a significant predictor of CVD events [hazard ratio 0.61, 95% CI (0.51–0.72) p <0.0001]. In similar analyses FMD remained significantly associated with incident CVD events after adjusting for 1.) age [hazard ratio 0.70, 95%CI (0.60–0.85), p=0.001], 2) age and gender [hazard ratio 0.79, 95%CI(0.65–0.97), p=0.010] 3.) age, gender, diabetes, smoking status, LDL, HDL, triglycerides, systolic BP, use of hypertension medications, resting heart rate and statin use (the full model), and 3) the Framingham Risk Score [hazard ratio 0.80, 95%CI(0.63–0.97), p=0.025] (). shows the hazard ratios of variables in the univariate (adjusted for age and gender) probability-weighted Cox model, the variables that made it into the multivariable model based on the a priori p value <0.20 and the hazard ratios of the variables in the multivariable probability-weighted Cox model.
Figure 1 Weighted curves showing cumulative event-free survival stratified by FMD > or =/ < the sex-specific median value for the cohort (follow-up truncated at 1700 days due to small numbers of subjects at risk for follow-up > 1700 days).*indicates (more ...)
Figure 2 Probability-weighted Cox Proportional Hazard Ratios for nested models of FMD vs CVD events with and without cardiovascular risk factors and the Framingham Risk Score (FRS). * of change in FMD/ unit SD. ** adjusted for age, gender, diabetes mellitus, cigarette (more ...)
Univariate (age and gender adjusted) and multivariable hazard ratios (probability weighted) and 95%CI for cardiovascular event incidence in relation to FMD and risk factors.
In stratified analyses, the inverse association between FMD and CVD events was similar across gender, smoking, hypertension and diabetes strata and there were no significant interactions between any of these predictors of CVD events and FMD in determining prognosis (data not shown). In a similar fashion, FMD was similarly associated with CVD events in all four ethnic groups; however, the confidence limits all included unity in fully adjusted models ().
Figure 3 Univariate* and multivariable**(full model) probability- weighted Cox proportional hazard ratios of change in FMD/unit SD vs CVD events stratified by ethnicity. Full model was adjusted for age, gender, diabetes mellitus, cigarette smoking, systolic blood (more ...)
Significant associations were also observed between FMD/unit SD and the major elements of the primary outcome in univariate probability-weighted Cox proportional hazard analyses; and, despite reduced sample size, FMD remained significantly associated with both MI and CVD death in fully adjusted models ().
Hazard ratio (95%CI) of FMD/unit SD for the primary outcome (CVD events) and its major constituents* in probability-weighted univariate (adjusted for age and gender) and multivariable models**.
In ROC analyses, the c statistic (AUC) for a univariate model of FMD was 0.65 while the c statistic for a model containing the Framingham risk score was 0.74. Addition of FMD to the Framingham risk score or to our full model did not increase the c statistic 0.74 (). Examination of the re-classification properties of FMD indicate that a risk model that adds FMD to the FRS net correctly re-classifies 52% of subjects with no incident CVD event, but net incorrectly reclassifies 23% of subjects with an incident CVD event. The overall net correct re-classification is 29% (p < 0.001 ). In the FRS intermediate risk subgroup the net correct re-classification was similar (28%, p < 0.001)
Receiver Operator Curves for the Framingham risk score (AUC=0.74), brachial FMD (AUC=0.65) and Framingham risk score + FMD (AUC=0.74) to predict incident CVD events.
Reclassification of subjects based on FRS+ FMD vs FRS alone
Brachial artery diameter (height adjusted), was a significant predictor of CVD event is the univariate probability weighted Cox analysis [hazard ratio 1.52(95%CI, 1.34–1.72), p<0.0001], after adjusting for FRS [hazard ratio 1.307(1.14–1.50), p=0.029] but was not an independent predictor of events in our final model [hazard ratio 1.13(95%CI, 0.75–1.63), p=0.59) ().
Hazard Ratio (95%CI)for cardiovascular event for brachial diameter/ unit SD (height adjusted) in univariate and four multivariable models.
The c statistic of brachial artery diameter was 0.64 and addition of brachial artery diameter also failed to increase the c statistic of the FRS (c statistics=0.74).