In this study of 3,769 men and women of White ethnicity, we found that several traditional cardiovascular risk factors as well as indicators of low-grade inflammation were associated with aortic stiffness 16 years later, and that the pattern of associations differed by sex. The strongest determinants of aortic stiffness, in order of magnitude, were waist circumference, waist-hip ratio, heart rate and IL-1Ra among men, and adiponectin, triglycerides, pulse pressure and waist-hip ratio among women.
To our knowledge this is the first study to examine the long term effect of traditional and novel cardiovascular risk factors on aortic stiffness in both men and women. In our study, central obesity and low-grade inflammation were strong predictors of aortic stiffness in both sexes. A marked difference between sexes was observed in the impact of triglycerides and adiponectin, which was twice as high in women, and in heart rate which was almost three times as high in men than in women.
Our findings among men are broadly in line with those from the Caerphilly study 
on 825 Welsh men during 20 years of follow-up. However, there were differences in the ranking of the determinants between the two studies, as low-grade inflammation was one of the strongest determinants of aortic stiffness and had a much larger impact than central obesity on aortic stiffness in the Caerphilly study. These differences may be attributable to the lower mean age and healthier risk profile of the men in the Whitehall II cohort and may also reflect the difference in level of adjustment in the analyses. Both studies adjusted for age and mean arterial pressure. We additionally adjusted for quadratic age, BMI and medication- and event history where relevant, whereas the Caerphilly study instead adjusted for heart rate and medication at time of follow-up only.
The observed sex differences in the relative importance of long-term determinants of aortic stiffness are important as it may point to differences in the etiology of cardiovascular disease between men and women. These findings also hold possible clues for how preventive strategies might be targeted differently by sex. It is, however, important to note that in the occupational Whitehall II cohort lower employment grades are overrepresented among women compared to men. This key indicator of socioeconomic status 
has shown to be strongly related to cardiovascular risk factors and general health in this cohort 
. Although we found no confounding effect of employment grade in our analyses, we cannot fully discount the role of the sex specific socio-economic structure in the Whitehall II population.
Determinants of aortic stiffness
Central obesity and inflammation
Waist circumference and waist-hip ratio were among the strongest predictors of aortic stiffness among men and women, but with somewhat higher effect in men. The association with waist circumference was strengthened upon adjustment for BMI indicating that central obesity is the main contributing factor to the association between obesity and aortic stiffness. Our findings are in accordance with a cross-sectional study measuring visceral fat by computer tomography in 2,488 men and women. The study found that besides systolic blood pressure, visceral fat had the strongest independent association with aPWV 
. The importance of fat accumulation as a determinant of aortic stiffness was further highlighted by our findings of a strong negative association between adiponectin and aortic stiffness in both sexes. Results
from cross-sectional studies are all in line with our results 
We also found other markers of low-grade inflammation, such as CRP, IL-6, IL-1Ra and fibrinogen to predict aortic stiffness in both sexes. Factor VII were additionally associated with aortic stiffness in men. These findings are in line with the Caerphilly study, which found an association of CRP and fibrinogen with aortic stiffness in men 
and several cross-sectional studies reporting a strong association between CRP and aPWV 
. There are only few studies on the association between IL-6, IL-1Ra, fibrinogen, von Willebrand factor and factor VII and aPWV, and the results are inconsistent 
Although the association between inflammation and aortic stiffness may merely reflect the inflammatory burden caused by aortic stiffness or its determinants, an experimental study showed that acute systemic inflammation induced by vaccination with Salmonella typhi
increases aPWV 
, supporting the hypothesis that inflammation is actually on the causal pathway leading to aortic stiffness. A further study supports this concept by showing that aortic stiffness is increased in people with rheumatoid arthritis, and that stiffness may be reversed by immunomodulatory therapy 
Dyslipidaemia is an important, well-established and modifiable cardiovascular risk factor. However, apolipoprotein A–I and apolipoprotein B may have an even stronger effect on coronary heart disease 
. We are the first to study the impact of a detailed lipid profile on aortic stiffness and found that HDL cholesterol and apolipoprotein B were associated with aortic stiffness in both sexes but with higher absolute and relative magnitude of apolipoprotein B in women ( and ). Additionally, LDL cholesterol in women and apolipoprotein A–I in men were to a lesser degree associated with aortic stiffness.
We found triglycerides to have a robust association with aortic stiffness in both sexes, but with an association twice as strong in women compared to men. This striking sex difference is also supported by a meta-analysis on incident coronary heart disease 
. The mechanisms behind this sex difference are poorly understood but may be a result of sex-specific differences in lipid metabolism. Our findings are in line with the findings of the Framingham Offspring Study, which identified apolipoprotein B and HDL cholesterol as major determinants of incident coronary heart disease 
Few studies exist on the long term effect of markers of glucose metabolism on aortic stiffness. In our study, the magnitude of the association between glucose metabolism and aortic stiffness after 16 years was comparable to that of inflammation and blood pressure in men but smaller than that of triglycerides and obesity. We found no association with aPWV in women. The Caerphilly study measured fasting plasma glucose as a marker of glucose metabolism but found no association with aortic stiffness.
Heart rate and blood pressure
There were also marked sex differences in the association between heart rate and pulse pressure and aortic stiffness. The association with heart rate was almost three times higher in men as in women, whereas pulse pressure was significantly associated with aortic stiffness in women only. The findings in the Caerphilly study on heart rate and pulse pressure are not in agreement with our results in men, which could be explained by the Whitehall II cohort being younger and thus pulse pressure may be a less accurate reflection of aortic stiffness.
The association between alcohol consumption and aortic stiffness differed by sex. Among men, we found a statistically significantly positive association with aortic stiffness. In women, however, the association was negative, although not statistically significant. This difference between sexes could be a reflection of social class, as the alcohol intake was markedly larger among highly educated women than among women with no education, whereas the difference in alcohol consumption across education level among men was less pronounced 
Strengths and limitations
This study included middle-aged British civil servants of White ethnicity limiting the generalizability of our results. Due to the lack of a baseline measure of aPWV, we cannot conclude on causality. Clinical trials targeting the identified risk factors for aortic stiffness are needed to assess causality. A tape measure was used to determine the carotid-femoral path length which may have overestimated the distance in obese individuals resulting in an overestimated aPWV.
For most of the determinants 5% or more were missing. Instead of complete case analysis, we have used multiple imputation to handle the large number of missing data in this study. Several simulation studies have shown that complete case analysis generally leads to biased estimates, but that multiple imputation reduces this bias and increases precision 
. MICE is currently the state-of-the-art-method of dealing with data missing at random 
Of the 7,955 participants attending the phase 3 examination, 3,894 had measurements of aPWV at follow-up. We cannot exclude bias due to the healthy survivor effect, which might have weakened the associations, but the ranking of the determinants should not be affected by this effect.
We cannot fully conclude on whether the found associations are truly predictive or merely an effect of the risk tracking over time. We have adjusted the analyses for medication history, incident diabetes and coronary heart disease to account for temporal instability in the predictor variables that was outside the natural course of ageing, and the ranking and strength of the associations were largely replicated in the cross-sectional analyses. However, future studies assessing the trajectories of risk factors up to the time of aortic stiffness measurement would enable a more precise quantification of the impact of the risk factors over time.
In conclusion, this large prospective study of middle-aged men and women found central obesity and low-grade inflammation to be strong predictors of aPWV in both sexes. In addition, heart rate in men and adiponectin and triglycerides in women were strongly associated with aPWV suggesting that prevention strategies targeting aortic stiffness should focus on central obesity and heart rate among men and triglycerides and central obesity among women.