An old water pipe hums differently from a new one. Once a pipe loses its flexibility, every push of water returns to the pump faster than in a young system. The same thing happens with our arteries: over the years their walls harden, and the blood wave, having reached the small vessels, reflects back to the heart almost without delay. This early return of the wave can be picked up by a sensor on the neck, and from it you can estimate the age of the vascular system.
Eight features were extracted from the resulting curve. The most important of them is the reflected-wave index VRI. It shows how large the reflected wave is compared with the main one. In young arteries the reflection is weak, because flexible walls damp the wave the way a soft pillow absorbs a blow. In older people the walls are stiff, and the reflected wave comes back almost as strong as the original. The correlation of VRI with age turned out to be the highest among all features: r = 0.738, and after adjustments for sex and exercise habits it rose to 0.798. This means that VRI is the most reliable marker of vascular age.
Before this work it was unclear which of the eight carotid Doppler features is most strongly linked to age and whether it would hold up after adjustments. Now it is clear: the reflected wave is the most reliable marker. It is measured with a simple portable sensor that can be used in a general practitioner's office or in the field. However, until VRI has been compared with reference measures of arterial stiffness, it is too early to call it a ready clinical tool. But even today it can be computed on portable devices for a preliminary assessment of vascular age.
The gap between the strongest and the weakest feature in the random forest was almost 20-fold: VRI received an importance of 0.536, while d-min got only 0.025. This shows how much more informative the reflected wave is compared with the diastolic minimum.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.