Think about that: They’re claiming to detect meaningful differences of 0.4 m/s when their baseline measurement uncertainty is 0.88 m/s. It’s like trying to measure the height difference between two people using a ruler that’s already bent by twice the amount you’re trying to measure.
INGA314.ai – How a powerful new analysis framework revealed 22 critical logical errors in prestigious cardiovascular research
https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehaf430/8236450
The Promise and the Problem
In August 2025, the European Heart Journal published the CARTESIAN study, a multicentre cohort spanning 2,390 participants that examined arterial stiffness 6–12 months after COVID‑19. After multivariable adjustment, COVID‑positive groups showed modestly higher carotid‑femoral pulse wave velocity (cfPWV) than controls; sex‑stratified analyses suggested the effect was concentrated in women.
A companion ESC press release adds operational texture: recruitment ran September 2020–February 2022 across 34 centres in 16 countries; cfPWV was measured roughly 6 months post‑infection and again at 12 months; vaccinated participants “generally had arteries that were less stiff,” and group means “stabilised or improved slightly” at 12 months.
Those are careful, interesting findings. The issue is what happens next: interpretation and extrapolation.
What INGA314.ai Is (and Why It Exists)
INGA314.ai is a logic‑audit framework for scientific claims. Think “spell‑check for reasoning.” It inspects:
Applied to CARTESIAN, the audit flags 22 issues across these domains. The study is valuable; the logic around it needs guardrails.
Small Effects, Big Interpretation: The Detectability Problem
What was reported
Adjusted cfPWV was higher by +0.41, +0.37, +0.40 m/s in COVID‑positive groups versus controls (adjusted control mean 7.53 m/s, 95% CI 7.09–7.97). In sex‑stratified analyses, women showed differences of +0.55, +0.60, +1.09 m/s across severity strata; men did not.
What that means for measurement
Detectability should be judged against minimal clinically important difference (MCID) or minimal detectable change for cfPWV under validated protocols—not against the width of a control confidence interval. Contemporary reproducibility work indicates that, in practice, ≈1.0–1.1 m/s is the range where within‑person changes can be distinguished from noise with reasonable confidence, especially when using optimal multi‑repeat protocols. The AHA/Hypertension validation guidance also treats ≤1.0 m/s device error as “good.” Together, those benchmarks make pooled ~0.4 m/s between‑group deltas borderline for device‑level detectability and clinical salience.
Why this matters
If effects sit on the measurement noise floor, centre/device heterogeneity, environmental conditions, and protocol variation can tip results. That doesn’t nullify the signal; it means interpretation must be modest until corroborated above threshold or anchored to outcomes.
Sex Differences: Real Heterogeneity Requires Real Modelling
The pooled effect can be significant even if a signal is concentrated in one subgroup. The right question is whether heterogeneity was fully modelled: a sex×COVID interaction, plus adjustment for menopausal status and HRT in women. That matters because midlife vascular stiffness trajectories accelerate around the menopause transition in many cohorts.
Framed correctly: a female‑skewed signal is plausible on background physiology, but mechanistic claims demand the corresponding covariates and interaction tests. Without those, pooled statements (or headlines) can mislead.
“Accelerated Ageing” Without a Baseline
CARTESIAN provides state differences at ~6 months and a second snapshot at 12 months; it does not provide pre‑infection baselines. That’s good cohort epidemiology, but it cannot show a change in rate (acceleration) without either pre‑COVID anchors or trajectories benchmarked against age/BP‑adjusted normative curves. The study reports partial stabilisation at 12 months in COVID‑positive groups and progression in controls, which is interesting and hypothesis‑generating, not a direct observation of accelerated biological ageing.
For context, global cfPWV averages around 7.45 m/s (95% CI ~7.11–7.79)—the CARTESIAN control mean sits right in that pocket—underscoring how fine‑grained these deltas are.
Vaccine‑Era Confounding: Measured ≠ Modelled
The press release notes lower PWV in vaccinated participants and implies vaccination was captured, but the abstract doesn’t show how vaccination entered the model (dose count, time since last dose), nor whether pre‑vaccine vs vaccine‑era infections were stratified—particularly by sex and severity. In a surrogate sensitive to selection and care‑seeking patterns, those distinctions matter.
INGA314.ai VEAE checks (applied):
- Era stratification (pre‑vaccine vs vaccine‑era infections)
- Dose and time‑since‑dose covariates
- Sex‑specific vaccination effects
- Baseline cardiovascular risk balance by vaccination status
If any are missing, flag confounding risk and request a stratified supplement.
Discussion Discipline: Keep Narrative Behind Data
The editorial tone is appropriately cautious—large multicentre cohort, interesting sex‑specific signals, and a call to identify modifiable targets and link surrogates to clinical outcomes. That’s the right slope: association → mechanism hypotheses → outcome validation. Claims about “accelerated vascular ageing” should be treated as working theory pending baseline‑anchored trajectories and event data.
The One‑Screen Logic Audit (journal‑friendly)
| Domain | Concrete check | CARTESIAN status | Why it matters |
|---|---|---|---|
| Measurement validity | Is ΔcfPWV > practical detectability (≈1.0–1.1 m/s) under protocol? | Pooled Δ ≈0.37–0.41 m/s; female ICU up to 1.09 m/s | Borderline detectability for pooled effects; heterogeneous robustness. |
| Effect heterogeneity | Sex×COVID interaction reported; menopause/HRT adjusted? | Sex‑stratified shown; interaction/HRT not in abstract | Prevents Simpson‑style misreads; physiology supports modifiers. |
| Temporal logic | Baselines or trajectories vs normative curves? | 6‑ and 12‑month snapshots; no pre‑COVID baseline | “Acceleration” is a rate, not a single state difference. |
| Vaccine‑era structure | Era/dose/time‑since‑dose, by sex/severity? | Vaccination associated with lower PWV; modelling granularity unclear in abstract | Era and selection can tilt surrogate markers. |
| Outcome linkage | Surrogate tied to hard events before clinical claims? | Not yet (cohort follow‑up planned) | Avoids correlation → causation inflation. |
What the Study Did Right
- Large, international cohort with standardised cfPWV protocols and two timepoints (6 and 12 months).
- Sex‑stratified reporting and explicit acknowledgment of uncertainty about clinical meaning.
- Public communication noting vaccination association and stabilisation over time.
These are strengths worth building on.
How to Say It Without Over‑Saying It
In this cohort, cfPWV—a surrogate of large‑artery stiffness—was modestly higher in female COVID‑19 survivors at ~6 months, with partial stabilisation by 12 months. Whether these small differences represent clinically meaningful vascular change remains unknown without outcome linkage and granular vaccine‑era modelling. Claims of “accelerated ageing” should be framed as hypotheses pending baseline‑anchored trajectories.
The Bigger Picture: Logical Hygiene as Standard Practice
We already police statistics, design, and data integrity. It’s time to systematise logic:
- Detectability boxes for any surrogate delta (state MCID/MDC and device validation context).
- Mandatory heterogeneity models for plausible modifiers (here: sex, menopause/HRT).
- Vaccine‑era stratification (era × dose × time‑since‑dose × sex).
- Temporal claims discipline (rates require baselines or validated normative trajectories).
- Outcome anchoring before clinical recommendations.
Do that, and the headlines will stop outrunning the evidence.
Conclusion: Science Deserves Better Logic
CARTESIAN advances the conversation: it documents small, sex‑skewed differences in a vascular stiffness surrogate months after COVID‑19 and shows hints of stabilisation by a year. It does not yet prove accelerated biological ageing or dictate clinical action. That’s not a failure—it’s a signal to tighten the logic and finish the story with baselines, stratified models, and, ultimately, clinical outcomes.
INGA314.ai exists to make that path explicit—so strong studies read as strong as they truly are.
About INGA314.ai
INGA314.ai is a system for identifying logical errors in scientific research, with extensions for vaccination‑era confounding and measurement validity.
Disclosure: Independent analysis; no funding or competing interests in COVID research.
Selected sources: CARTESIAN abstract (EHJ, Aug 2025); ESC press release; EHJ editorial; PWV detectability/reproducibility (J Clin Med 2023; AHA/Hypertension validation guidance); global cfPWV norms; menopause‑related stiffness trajectories.
Post‑publication note for readers who like the weeds: cfPWV is a surrogate for large‑artery stiffness with population‑level prognostic value; a ~0.4 m/s cross‑sectional delta at 6–12 months sits near the operational noise floor for many protocols. That’s why outcome linkage—not just statistical significance—should be the finish line.
System D 