Among all the CMR parameters, only LVM and LVM index (LVMI) changed significantly according to baseline BNP values (Online Table 3). Correlation analysis showed a strong linear relationship (r = 0.71, p < 0.01) between ΔLVM and baseline BNP levels (Figure 1) There was a strong positive relationship Proteasome inhibitor between BNP tertiles and ΔLVM (Figure 2, Online Table 3). This was the case whether the BNP tertiles were calculated on the basis of the tertiles of the original study (n = 300) or the tertiles of this substudy

(n = 50) (Figure 2). It is worth noting that the difference between tertile 1 and tertile 3 is large at nearly 12% of the mean baseline LVMI. The independent predictive value of the interaction between ΔLVM and baseline BNP levels for explaining the evolution of LVM with time (dependent variable)

was investigated by multiple linear regression analysis. We investigated 5 different models (Table 3). Model 1 was composed of previously reported clinical predictors of LVM such as age, sex, BP, body mass index, and history of smoking. Subsequent models explored the additional predictive value of adding total cholesterol and uric acid and then adding baseline hs-TnT or BNP on top of model 1. As shown in Table 3, both hs-TnT and BNP offered additional predictive value when added to the model by improving the c-statistics significantly. In a logistic regression analysis, BNP stood out as a strong predictor of a future rise in LVM. A receiver-operating characteristic analysis yielded a c-statistic of 0.88 for BNP with

a sensitivity and specificity of 70% and 88%, respectively, OSI 906 Amino acid at a BNP level of 17 pg/ml. Our main finding is that, in well-controlled primary prevention patients, a high BNP in the absence of any cardiac abnormality is able to identify those individuals whose LVM will increase during the next 3 years, that is, an elevated BNP is able to predict future increases in LVM. This may partly explain why, in so many studies, BNP predicts prognosis independent of echocardiographic abnormalities. The Framingham study has already shown that the tendency for LVM to increase with aging in a population is highly variable from one individual to the next 11 and 12. Increases in LVM in treated hypertension are, however, far from innocent (13). Serial changes in LVM predict CV events, independent of baseline LVM and independent of baseline BP or the degree of BP reduction (14). It now appears from our data that BNP can identify those whose serial LVM will increase with time, and we know that such individuals are at increased risk and that they are currently inadequately identified by either baseline LVM or any BP parameter (14). A major strength of our study is that the population studied was comprehensively phenotyped at baseline, that is, they were all assessed for LVM, LV systolic dysfunction, left atrial enlargement, LV diastolic dysfunction, and most importantly, for silent myocardial ischemia.