Does a Younger “Vascular Age” Mean Improved Vascular Structure?

A younger predicted age or lower pulse wave velocity alone does not establish structural rejuvenation. Pressure, smooth muscle tone, calibration, and an algorithm's actual-age input can contribute to the reading. At the same time, PWV has genuine prognostic evidence and favorable changes during long-term treatment; those findings should be retained.

01What is being measured?

Pulse wave velocity (PWV) is the speed of a pressure wave, not blood flow. Faster waves generally indicate greater effective stiffness, but stiffness depends on wall material, thickness, diameter, smooth muscle tone, and the pressure distending the vessel at measurement. Mechanical overview Primary pressure theory

A baPWV around 14 m/s and a cfPWV around 8 m/s cannot be subtracted to calculate years of rejuvenation: their paths and measurement methods differ. Dataset dictionary Human measurement methods

02Predicting birthdays: how much vascular health has been validated?

The 2026 Chinese descriptor provides 2015–2022 routine examination data from one Xiangyang hospital, ages 30–80, including sex, bilateral baPWV, pressure-related fields, ABI and a binary Label. Its regression target is chronological age, not imaging, histology or subsequent clinical events. Paper Pinned data

The original paper reports mean age errors across 20 random splits: in men, 7.58 years for multiple linear regression, 7.09 for SVR and 6.53 for KDM; in women, 6.37, 6.03 and 5.48. Better nonlinear performance should be acknowledged. However, 20 splits of one cohort are not replication in 20 populations. Its bootstrap intervals concern the mean of repeated split results, not guaranteed external performance. Table 3 and statistics Locked code

After auditing the workbook, we used the same 20 split seeds per sex for prespecified age-blind linear models. The training/validation/test proportions were approximately 70/15/15%; validation was reserved but unused for tuning. Mean test absolute errors, in years, were:

The fixed six are bilateral PWV, PMap and ankle mean pressure. This is a declared diagnostic comparison, not a rerun of each author's SHAP selection or the full seven-model pipeline. The results show age information in these measurements. They do not validate treatment-induced wall changes or reconstruct missing comorbidity, smoking or medication information.

Short lines span the minimum to maximum of 20 split results, not clinical-effect intervals. The last two rows share a fitted KDM but differ in whether the test person's actual age is supplied. Their prediction tasks differ.

An actual-age term that is easy to miss

KDM can use age as a prior in a biological-age estimate. That declared construction is not inherently illegitimate. Ranking it by birthday error against models that do not know the birthday, however, requires acknowledging the input difference. The locked code supplies actual test age to KDM; Age is absent from the six learning models' features. KDM and main code

Writing our implementation in simplified form, M is the biomarker-only estimate from the same fit, CA is actual age, and w is determined from training data:

KDM = (1 − w) × M + w × CA
KDM − CA = (1 − w) × (M − CA)

For that fixed fit, providing CA therefore contracts each person's age error by the same factor. With our fixed six, mean w was 0.453 in men and 0.365 in women. Biomarker-only KDM mean errors were 12.76 and 9.32 years; with CA they became 6.98 and 5.92. No healthier wall is needed to explain this mathematical change. Nor does it show that the remaining age difference lacks prognostic information.

Our 6.98/5.92 are not a full reproduction of the published 6.53/5.48, and the original ranking advantage cannot all be attributed to this term. Changing inputs changes the fit: with all ten, our age-supplied KDM errors were 7.90/6.78, exceeding the corresponding age-blind OLS errors of 7.27/6.27. Algorithm construction, birthday error and biological validity need separate assessment.

03A younger age difference may include calibration bias

An age-blind predictor can distinguish ages while producing a narrower distribution than actual ages. In the first prescribed test split, fixed-six OLS overestimated men aged 30–39 by an average 11.05 years and underestimated those 70–80 by 11.54 years. Corresponding errors in women were +8.14 and −10.37 years.

This is a cross-sectional calibration check, not an individual's aging rate. Lower predicted age in an older group is not direct evidence of protection. Age-supplied KDM also contracts differences by construction. Repeated testing requires a fixed model, preprocessing and reference, alongside repeatability and within-age outcome validation. Calibration can reduce distributional bias without turning the output into a directly measured tissue age.

The data audit leaves three specific boundaries. The 13 released columns contain no missing values to impute, so a description of median imputation before splitting does not establish material leakage in this version; earlier collection/curation remains unknown. There are two male left-baPWV values of 99.99 whose meaning is unconfirmed. Removing only flagged training rows while keeping the same splits and entire test sets changed fixed-six male OLS MAE by less than 0.03 years. This does not validate the extreme readings physiologically.

The paper calls PMap brachial mean pressure in mmHg, whereas the official ARIC codebook uses the same field names for percentage mean BP of the waveform and lists actual brachial mean BP separately. This clue cannot definitively relabel the Chinese instrument export, but prevents confidently interpreting a PMap near 54 as 54 mmHg. We retain PMap as an unresolved numeric input and show models excluding it. Official codebook

Published means differ from this pinned release by less than 0.01, but one SD difference exceeds ordinary display precision: male right PMap is reported as 3.26 and recalculates to 3.226. The cause is unconfirmed. We retain the original report and do not equate calculation consistency with valid measurements or original inputs.

The paper's binary Label definition also lacks the right-PWV comparator. Inserting the intuitive “≤15.2” while retaining the other two conditions disagrees with 6,234 released labels. None of the other frozen candidates fully reconstructed Label either. Higher agreement does not establish the author's intended rule. Label was excluded from both age regressions and is not an independently followed clinical event. These uncertainties do not invalidate every age prediction. Table 7

04Pressure can lower wave speed without changing the wall law

A vessel does not behave identically at every pressure. Changing the working point can change which wall components bear the load. We examined one specific explanation using the exponential pressure–diameter law of the primary theoretical study: Equations and assumptions

P = P0 × exp[β × (D / D0 − 1)]
c² = (A / ρ) × dP/dA
   = P × [β + ln(P/P0)] / (2ρ)

P is transmural working pressure, D diameter, A cross-sectional area, ρ blood density and c this model's local wave speed. Calculations convert mmHg to Pa. We chose P0=100 mmHg, D0=6 mm, ρ=1050 kg/m³, and β=10. These are not fits to the Chinese participants, and ordinary recorded BP is not automatically this segment's transmural pressure.

On this unchanged wall law, lowering pressure from 100 to 80 mmHg lowers wave speed from 7.97 to 7.05 m/s. At 100 mmHg, choosing β≈7.821 also yields 7.05 m/s. Thus, a lower reading cannot uniquely identify a changed wall law within this model.

These are finite examples using chosen parameters, not treatment results. β is an effective parameter of a simplified law, not collagen content, elastin content or histological age. A complete pressure–diameter relation need not be a single exponential. Matching pressure or applying correction can sharpen the question but still requires model and measurement validation. Adding pressure to a regression does not automatically separate immediate distension, tone and long-term remodeling.

Wave-foot propagation measurements usually concern the diastolic working point. The plot's local transmural pressure therefore cannot be arbitrarily replaced with one recorded mean BP as a universal clinical correction. Diastolic point in the theory Human pressure-correction methods

05A human counterexample: BP and PWV need not move together

A 2022 study made acute measurements before and after nitroglycerin in 40 Veterans over 60 without known cardiovascular disease, 20 hypertensive and 20 nonhypertensive. During carotid assessment, group mean pressures fell 93→82 and 97→88 mmHg, while derived local carotid PWV rose 6.0→6.6 and 6.8→7.8 m/s. Measured cfPWV also rose; local brachial PWV showed no clear change. Tables 2–4

This provides another mechanism clue: smooth muscle relaxation, diameter and load sharing need not reduce effective stiffness in every vessel type. Some increases persisted after the authors' exponential pressure correction. This does not establish damaged extracellular matrix within minutes or convert an acute PWV change into an event hazard.

Important limits remain. Six control brachial scans were excluded, leaving 34 at that site. Post-treatment local values came from the two maximum-diameter time points; cfPWV used the two largest-change time points, not average responses over the experiment. Peripheral BP was used for local carotid calculations, heart rate was unadjusted and modeled component separation had no histological validation. Some abstract/text P values do not align with table columns; we use explicitly located table results and retain the discrepancy. This mechanistic experiment provides no recommendation to start, stop or adjust medication.

06Prognostic value does not establish a surrogate endpoint

A 2014 individual-participant meta-analysis covered 16 prospective studies, 17 cohorts and 17,635 people, with 1,785 cardiovascular events. After age, sex, BP and other risk-factor adjustment, a one-cohort-SD increase in log aortic/cfPWV corresponded to cardiovascular-event HR 1.30 (95% CI 1.18–1.43). That supports prognosis; it cannot be directly transferred to baPWV, acute changes or our age model. Methods and Table 1

Adding PWV improved cardiovascular-event C-index by about 0.0052 (0.0011–0.0092), a modest discrimination increment. Reclassification used study-specific quantile cutoffs derived from event participants' predicted risks; it does not establish an unconditional 13% clinical benefit at conventional thresholds. Tables 2–3

SPARTE then randomized management informed by PWV versus conventional BP management. The original flow shows 544 randomized and 536 in the main analysis, 264 versus 272; eight were excluded for eligibility or no follow-up. Over a median 48.3 months, primary cardiovascular/renal composite events were 17 versus 24, HR 0.74 (0.40–1.38), p=0.35. The interval allows substantial benefit and some harm. Recruitment fell short of its plan: neither proven benefit nor proven ineffectiveness follows. Flow, Table 2 and discussion

Favorable measurement results also matter: annual PWV increase was about 0.06 versus 0.20 m/s, with different trajectories even after including time-varying mean BP. But treatment intensity and BP changed too. Statistical adjustment is not direct observation of matrix remodeling and does not prove PWV mediation of clinical effects. The available registration differs in planned enrollment and was first posted after recruitment began; it is not fully aligned prospective registration. Registry

A testable assessment begins with a fixed measurement, pathway and model; uses paired observations to account for concurrent pressure, heart rate and tone; examines pressure–diameter, geometry, thickness or material evidence for the structural hypothesis; and finally validates whether changes reliably predict clinical benefit. BP control can have genuine benefits through reducing load without demonstrating histological rejuvenation first. An age number needs these bridges to carry a stronger health claim.

Reproducibility archive: code, input facts and complete diagnostic results. It provides a pinned official-data retrieval method, all 33 original mean/SD comparisons, every input group, splits, parameters, label candidates and extreme-value sensitivity. We did not rerun the original seven models/SHAP or clinical IPD Cox/mediation analyses, and did not obtain paired longitudinal wall structure and outcomes for these examination participants.