Ferritin and transferrin predict common carotid intima-media thickness in females: a machine-learning informed individual participant data meta-analysis

  1. Anand Ruban Agarvas
  2. Richard Sparla
  3. Janice L Atkins
  4. Claudia Altamura
  5. Todd Anderson
  6. Ebru Asicioglu
  7. Judit Bassols
  8. Abel López-Bermejo
  9. Hana Marie Dvořáková
  10. José Manuel Fernández-Real
  11. Christoph Hochmayr
  12. Michael Knoflach
  13. Jovana Kusic Milicevic
  14. Silvia Lai
  15. José María Moreno-Navarrete
  16. Dariusz Pawlak
  17. Krystyna Pawlak
  18. Petr Syrovatka
  19. Dorota Formanowicz
  20. Pavel Kraml
  21. Jose M Valdivielso
  22. Luca Valenti
  23. Martina U. Muckenthaler

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Background

Iron overload promotes atherosclerosis in mice and causes vascular dysfunction in humans with Hemochromatosis. However, data are controversial on whether systemic iron availability within physiological limits affects the pathogenesis of atherosclerosis. We, therefore, performed an individual participant data (IPD) meta-analysis and studied the association between serum iron biomarkers with common carotid intima-media thickness (CC-IMT); in addition, since sex influences iron metabolism and vascular aging, we studied if there are sex-specific differences.

Methods

We pooled the IPD and analysed the data on adults (age≥18y) by orthogonal approaches: machine learning (ML) and a single-stage meta-analysis. For ML, we tuned a gradient-boosted tree regression model (XGBoost) and subsequently, we interpreted the features using variable importance. For the single-stage metaanalysis, we examined the association between iron biomarkers and CC-IMT using spline-based linear mixed models, accounting for sex interactions and study-specific effects. To confirm robustness, we repeated analyses on imputed data using multivariable regression adjusted for key covariates identified through machine learning. Further, subgroup analyses were performed in children and adolescents (age<18y). In addition, to evaluate causality, we used UK Biobank data to examine associations between the hemochromatosis (HFE) genotypes (C282Y/H63D) and mean CC-IMT in ∼42,500 participants with carotid ultrasound data, using sex-stratified linear regression (adjusted for age, assessment centre, and genetic principal components).

Results

We included IPD from 21 studies (N=10,807). The application of the ML model showed moderate predictive performance and identified iron biomarkers (transferrin, ferritin, transferrin saturation, and iron) as key features for IMT prediction. Multivariable analyses showed non-linear sex-specific relationships for ferritin and transferrin with CC-IMT: ferritin showed a significant positive association, and transferrin showed negative associations at specific ranges, both only among females. No significant associations were found between CC-IMT in those with HFE genotypes in either sex in the UK Biobank.

Conclusion

Our observational data show that iron biomarkers - ferritin and transferrin are non-linearly associated with CC-IMT specifically in females, while a significant causal association between the HFE genotype and CC-IMT could not be demonstrated in the UK Biobank data. We conclude that the observational associations may not only be explained by causal effects of iron on the arterial wall thickness, but also in part be driven by residual confounding factors such as inflammation.

Other: No financial support was received for this meta-analysis. The protocol for this study is registered in the PROSPERO database (CRD42020155429; https://www.crd.york.ac.uk/ ).

Article activity feed

  1. PREreview

    This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/16999814.

    Carotid intima-media thickness (CC-IMT) lacks universally accepted measurement protocols and normal reference values, undermining its use as a reliable, comparable surrogate across diverse cohorts. This variability is linked to inconsistencies in anatomical site, measurement methodology, and population differences.

    CC-IMT increases may reflect medial hypertrophy rather than atherosclerotic intimal thickening, especially in cases of hypertension, confounding interpretation regarding vascular pathology. The broader clinical community has raised concerns about CC-IMT's predictive value. In many settings, including routine practice, its utility is debated or considered marginal. So, if the CC-IMT metric is noisy, nonspecific, or only loosely tied to clinical outcomes, any purported relationship with iron biomarkers (like ferritin or transferrin) could be spurious or clinically irrelevant.

    The study focused on CC-IMT but excluded plaque presence (>1.5 mm). However, IMT and plaque represent different pathological processes. Failure to distinguish between adaptive vascular remodeling and early atherosclerotic changes further weakens the interpretive clarity.

    The meta-analysis aggregated CC-IMT data from 21 studies, but ultrasound protocols, operator expertise, and imaging equipment likely varied substantially. This introduces measurement heterogeneity and betrays the assumption of comparability across datasets.

    Finally, without incontrovertible validation of CC-IMT's relevance and consistency, the reported associations are at high risk of being artefactual. The complex modeling applied, machine learning, and spline regression cannot salvage findings built on an unreliable foundation.

    Competing interests

    The author declares that they have no competing interests.

  2. Version published to 10.1101/2025.04.30.25326720 on medRxiv