In patients with heart failure (HF) and reduced ejection fraction (HFrEF), iron deficiency (ID) is a known predictor of adverse outcomes (Jankowska et al., 2010; Klip et al., 2013). As iron is a key factor in erythropoiesis, energy metabolism, and mitochondrial function, disturbances in iron metabolism also impact functional status, including exercise capacity (Okonko et al., 2011). The diagnosis of ID in HFrEF is crucial, since intravenous correction of ID improves exercise performance, symptoms, and quality of life and might reduce hospitalizations, regardless of the presence of anemia (Anker et al., 2009; Ponikowski et al., 2015, 2020).

Less is known about the influence of iron status on functional status and myocardial performance in HF with preserved ejection fraction (HFpEF). HFpEF accounts for half of the HF hospital admissions, and its pathophysiology is characterized by both cardiac (e.g., diastolic dysfunction) and non-cardiac (e.g., skeletal myopathy) abnormalities causing exercise intolerance (Shimiaie et al., 2015; Gevaert et al., 2019; Pugliese et al., 2019). Observational data show a strong relationship between ID and a reduced aerobic exercise capacity, measured as peak oxygen uptake (peak V̇O₂), in patients with HFpEF (Beale et al., 2019). In addition, ID affects immune responses and oxidative stress mechanisms underlying these cardiac and non-cardiac abnormalities (Macdougall et al., 2012; Paulus and Tschöpe, 2013). In vitro, ID is associated with impaired mitochondrial respiration and reduced contractility and relaxation in human cardiomyocytes (Hoes et al., 2018). However, a relationship of ID to either exercise capacity or left ventricular (LV) stiffness could not be found in a small study of patients with HFpEF (Kasner et al., 2013).

While several drugs are able to reduce HF hospitalizations, a therapy that improves mortality in HFpEF is still lacking (Gevaert et al., 2022). A key reason for this could be the underlying phenotypic heterogeneity of HFpEF (Shah et al., 2016). Subdividing and classifying patients with HFpEF by different disease phenotypes may increase our understanding of HFpEF pathophysiology, simplify its diagnosis, and allow for targeted management instead of a “one size fits all” approach (Shah et al., 2016; Gevaert et al., 2020).

Artificial intelligence can aid in this classification by recognizing patterns in large datasets. In unsupervised machine learning, relationships are determined based on raw data, independent of the existing classification. One approach is to find “clusters” of similar data items: subjects in the same cluster are similar to each other and dissimilar to subjects in other clusters. Previously, machine learning has been able to identify clusters of patients with different prognoses in patients with HFrEF and/or HFpEF (Kao et al., 2015; Shah et al., 2015; Hedman et al., 2020).

In this study, we assessed the iron status of patients with HFpEF included in the Optimizing Exercise Training in Prevention and Treatment of Diastolic Heart Failure (OptimEx-Clin) multicenter trial (Mueller et al., 2021). We aimed to establish the relationship of iron parameters with diastolic function and aerobic exercise capacity, and the contribution of iron status to patient phenotyping using machine learning.

We concluded that in patients with HFpEF, (i) ID is a very frequent comorbidity, (ii) iron parameters relate to diastolic function and aerobic exercise capacity only in patients with ID, (iii) patient phenotyping is significantly impacted by including ID, and (iv) a phenogroup of younger women with HFpEF has a high prevalence of ID but a few other comorbidities. Together, this forms evidence of a clinically important link between HFpEF and ID in some but not all patients.

The guideline-recommended management of HFpEF is still based on treating symptoms and controlling comorbidities (Yancy et al., 2017). In this study, we confirmed that ID is an important comorbidity in HFpEF. Affecting 59% of this multicenter European HFpEF population, ID is more prevalent than obesity, diabetes mellitus, CAD, chronic kidney disease, or atrial fibrillation. Previous reports in smaller HFpEF populations found a comparable prevalence of ID, ranging from 57 to 70% (Kasner et al., 2013; Núñez et al., 2016; Martens et al., 2018). ID seems more common in HFpEF than HFrEF, where a prevalence of 37–50% is reported (Jankowska et al., 2010; Okonko et al., 2011; Klip et al., 2013; Grote Beverborg et al., 2018). This could be explained by the larger proportion of women among patients with HFpEF, as female sex is also a known risk factor for ID (Beale et al., 2018).

Indeed, in our study, women had a threefold higher risk of ID compared to men. The cause of this sex difference is not clear, as ID in women is traditionally attributed to blood loss during menstruation and higher iron requirements during pregnancy (Beale et al., 2018), but all women in our study were postmenopausal. Other risk factors shared by HFpEF and ID (e.g., age, chronic kidney disease, and obesity) did not differ between patients with and without ID in our study.

In multivariable regression, the relationship between iron parameters and peak V̇O₂ was minor. This confirms findings from a recent analysis in 300 patients with HFpEF (Barandiarán Aizpurua et al., 2021). Previously, smaller studies found more important relationships between iron parameters and peak V̇O₂ (Núñez et al., 2016; Martens et al., 2018). Patients in these studies had a much lower exercise capacity than the current population, so disease severity is possibly a factor to consider in the relationship between iron status and aerobic exercise capacity. However, several other CPET parameters showed unfavorable changes in patients with ID: steeper VE/VCO₂ slope, lower peak O₂ pulse, and lower V̇O₂ at aerobic threshold (Table 1). These parameters have all been associated with the prognosis of patients with HF (Guazzi et al., 2012).

A possible explanation for these relations is the influence of ID on the O₂ cascade. The Fick principle states that V̇O₂ is the product of maximal cardiac output and arteriovenous oxygen extraction (Ca-vO₂). Both were impaired in patients with ID in a study combining CPET with echocardiography (Martens et al., 2021). Ca-vO₂ is determined by lung O₂ diffusion, peripheral oxygen extraction, and O₂ carrying capacity of Hb (Poole et al., 2018). Iron is also a crucial cofactor in mitochondrial oxidative phosphorylation (Gevaert et al., 2019). Thus, ID can reduce cardiac output through impaired mitochondrial function in cardiac myocytes, as well as Ca-vO₂ through reduced O₂ carrying capacity of Hb and impaired skeletal myocyte mitochondrial function.

This is the first study to demonstrate a significant association between iron parameters and diastolic function. In an observational study including 15 patients with HFpEF and ID, Kasner et al. (2013) could not find a relationship between iron parameters and invasive diastolic function. In our study, the relationship was weak after adjustment for age and sex in a multivariable linear regression model. However, when stratifying the population for the presence of ID, the relationship between diastolic function and iron parameters proved to be stronger in patients with ID.

From these data, it may be hypothesized that in patients with ID, iron depletion directly or indirectly impairs diastolic function. Mechanistic animal studies provide further insight in this relationship. Inducing severe ID-anemia in healthy rats directly caused diastolic dysfunction and HFpEF, associated with cardiomyocyte hypertrophy and fibrosis, as well as lung edema (Naito et al., 2009). In another study of the Dahl salt-sensitive rats, animals fed on high-salt diet rats developed not only HFpEF but also ID and mild anemia (Naito et al., 2011). This was associated with a downregulation of hepcidin, but in sharp contrast to ID without HFpEF induced by iron-depleted chow with normal salt content, this was related to a reduced duodenal iron absorption (Naito et al., 2011). Furthermore, iron-depleted cardiomyocytes have reduced mitochondrial function, causing a reduction in diastolic function (up to 85%) which is more important than the reduction in systolic function (up to 64%) (Hoes et al., 2018). Combined with our observations that ID relates to diastolic function and aerobic exercise capacity in patients with ID, these findings suggest that ID could play a more direct role in modulating diastolic function than previously thought. However, the cross-sectional nature of our study does not allow to separate cause and effect. Longitudinal studies and intervention studies, such as the upcoming effect of IV iron in patients with HFpEF (FAIR-HFpEF) randomized clinical trial, will provide more insight in the putative causal role of ID in HFpEF.

The cluster analysis provided another piece of this puzzle. When machine learning categorized patients with HFpEF in an unbiased way, three phenogroups were identified. The comorbidity profile of two of these phenogroups is familiar to many clinicians. The first can be summarized as atrial fibrillation + chronic kidney disease (AF-CKD) and the second as male + coronary artery disease (M-CAD). Similar phenogroups have been identified among other HFpEF populations subjected to machine learning analysis. Kao et al. (2015) first applied clustering in HFpEF using 11 features of patients in two randomized clinical trials. Six subgroups were discovered among which the AF-CKD and M-CAD groups can be recognized from this study. Shah et al. (2015) studied 46 features from a single-center HFpEF population. They could discern three phenogroups, including an AF-CKD population, whereas sex and CAD were more evenly distributed across subgroups. More recently, Hedman et al. (2020) found that mortality differed between six phenogroups in a European HFpEF cohort.

We extended this existing literature by incorporating iron parameters into a clustering analysis for the first time. This resulted in a phenogroup of predominantly female patients, who had a comparatively low prevalence of traditional HFpEF risk factors, such as advanced age, hypertension, obesity, CAD, and atrial fibrillation (phenogroup 1). These patients did have a remarkably high prevalence of ID of 86% (p < 0.001 vs. other phenogroups). Of note, Kao et al. (2015) and Shah et al. (2015) previously found phenogroups with comparable low-risk profiles, but could not link this to ID as iron status was not determined.

This low risk + iron deficiency (LR-ID) phenogroup shows a significantly higher peak V̇O₂ than the other phenogroups, which seems to contradict the relationship of ID with worse peak V̇O₂. However, patients in the LR-ID phenogroup were also younger, and the percentage of predicted peak V̇O₂ (more suitable to compare patients of different age distribution) was similar to other phenogroups.

Our findings have important therapeutic implications, as ID is readily treatable by oral or intravenous supplementation (Jankowska et al., 2013). In patients with HFrEF, the latter is preferred after randomized trials demonstrated that intravenous ferric carboxymaltose improved aerobic exercise capacity, symptoms, and quality of life in patients with HFrEF, whereas oral iron failed to improve symptoms (Anker et al., 2009; Ponikowski et al., 2015, 2020; Lewis et al., 2017). Mechanistic studies showed that in HFrEF, ID manifests as both myocardial and skeletal muscle iron depletion; however, only myocardial iron was repleted by intravenous iron (Melenovsky et al., 2018; Núñez et al., 2020). Hopes are high that future studies, such as the ongoing FAIR-HFpEF trial, will demonstrate similar effects of iron administration in HFpEF.

Our study has some limitations. The observational nature of the study precludes definite conclusions about causation. OptimEx-Clin was not powered for clinical outcomes, and together with the cross-sectional design of our study, this did not allow a prognostic assessment based on phenogroup membership. The phenogroup analysis should be validated in an external population before the clinical application of our findings. Finally, the exercise intervention in OptimEx-Clin may have induced a selection bias toward less symptomatic patients with HFpEF, filtering out those who are unable to exercise.

ID is a frequent and important comorbidity in patients with HFpEF. Lower iron parameters are associated with worse diastolic function and worse aerobic exercise capacity but only in patients with ID. Patient phenotyping by means of machine learning is significantly impacted by including ID. We identified a phenogroup of younger female patients with HFpEF with a high prevalence of ID and a few other HFpEF risk factors. Therefore, we suggest that iron status is routinely checked in all patients with HFpEF, as those with ID should be counseled on possible treatment options.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethisch Comité UZA, Antwerp University Hospital. The patients/participants provided their written informed consent to participate in this study.

AG, UW, BP, VA, MH, and EV contributed to the conception and design of the work. AG, SM, EW, AD, CV, EP-K, PB, and FE contributed to the acquisition, analysis, and interpretation of data for the work. AG drafted the manuscript. All authors critically revised the manuscript, gave final approval and agreed to be accountable for all aspects of work ensuring integrity and accuracy.

AG reports travel and accommodation funding by Vifor Pharma. EW reported receiving personal fees from Novartis (honoraria for lectures and advisory board activities), Boehringer Ingelheim (honoraria for advisory board activities), and CVRX (honoraria for lectures) outside the submitted work. AD reported receiving grants from Novartis outside the submitted work. CV reported receiving personal fees from Abbott, Daiichi-Sankyo, Bayer and Edwards Lifesciences (lectures) outside the submitted work. BP reported receiving personal fees from Bayer Healthcare (steering committee, lectures), Merck (steering committee, lectures), Novartis (steering committee, lectures), Servier, AstraZeneca (lectures), Bristol-Myers Squibb (lectures), and Medscape (lectures) outside the submitted work. MH reported receiving grants from Novartis (principal investigator of the Activity Study in HFrEF) and personal fees from Bristol-Myers Squibb, Berlin Chemie-Menarini, Novartis, Daiichi-Sankyo, AstraZeneca, Roche, Abbott (advisory board on exercise and diabetes), Sanofi, Pfizer, Boehringer Ingelheim, and Bayer outside the submitted work. EV was supported by an investigator-initiated grant from Vifor Pharma for this work. No other disclosures were reported. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.