Skip to content
Frontiers
Search

Original Research

Frontiers in Nutrition ·

Volume 13 - 2026 | https://doi.org/10.3389/fnut.2026.1985953

Laboratory-derived inflammatory–nutritional exploratory phenogroups and their associations with functional limitation and 30-day same-center all-cause unplanned readmission in patients with heart failure

  • Yilian Zhang1
  • Xu Zhang2
  • Hongbin Song3
  • Qiyu Hu1
  • Jingzhu Nan1*corresponding author
  1. 1Department of Clinical Laboratory, Beijing Anzhen Hospital, Capital Medical University, Beijing, China
  2. 2Department of Clinical Laboratory, The First Medical Center of Chinese PLA, Beijing, China
  3. 3Department of Clinical Laboratory, The Third Medical Center of Chinese PLA, Beijing, China
Article metrics0 viewsView details
PDFCiteMetrics

Abstract

Heart failure (HF) is characterized by marked clinical heterogeneity, while conventional left ventricular ejection fraction (LVEF)-based classification may not fully capture systemic inflammatory, nutritional, hematologic, and cardiorenal abnormalities. We aimed to identify laboratory-derived inflammatory–nutritional exploratory phenogroups in hospitalized patients with HF and evaluate their associations with functional limitation and 30-day same-center all-cause unplanned readmission. This single-center retrospective cohort included 1,588 hospitalized patients with HF, of whom 1,455 with complete data for 15 prespecified clustering variables were included in the primary analysis. K-means, Gaussian mixture models, and DBSCAN were compared, and K-means with K = 3 was selected based on internal separation, stability, cluster size, and clinical interpretability, with K = 2 used as a sensitivity analysis. Severe functional limitation was defined as NYHA class III–IV, and 30-day same-center all-cause unplanned readmission was evaluated as the short-term clinical outcome. Multivariable logistic regression with HC3 robust standard errors was used to assess these associations, with multiple-imputation and median-imputation sensitivity analyses conducted in the full cohort. Three phenogroups were identified: an inflammatory–cardiorenal–nutritional vulnerability phenotype, an anemia–immune dysregulation phenotype, and a relatively preserved laboratory phenotype. LVEF was similar across groups, whereas severe functional limitation and 30-day same-center all-cause unplanned readmission differed significantly. After enhanced adjustment, compared with Cluster 3, the odds ratios for severe functional limitation were 3.814 and 5.594 for Clusters 1 and 2, respectively, and the corresponding odds ratios for 30-day same-center all-cause unplanned readmission were 1.966 and 2.616. Higher quartiles of both ALI and PNI were associated with lower odds of severe functional limitation; higher ALI quartiles and the highest PNI quartile were associated with lower odds of 30-day same-center all-cause unplanned readmission. Routine admission laboratory measurements identified three exploratory HF phenogroups with distinct inflammatory, nutritional, hematologic, and cardiorenal profiles. After adjustment for continuous LVEF and NT-proBNP, the two vulnerable phenogroups remained associated with severe functional limitation and 30-day same-center all-cause unplanned readmission, with directionally consistent findings in the full-cohort and K = 2 sensitivity analyses. ALI and PNI reflect laboratory-derived inflammatory–nutritional vulnerability but cannot replace formal nutritional assessment or establish incremental predictive value. These findings remain exploratory and require validation in prospective multicenter cohorts.

Keywords

1 Introduction

Heart failure is a common clinical syndrome representing the advanced stage of multiple cardiovascular diseases and an increasing public health burden in aging populations. Global burden-of-disease studies indicate that the number of individuals living with HF continues to rise, with substantial geographic and socioeconomic disparities (1). Contemporary guideline-directed pharmacologic therapy, device therapy, cardiac rehabilitation, and multidisciplinary management have improved clinical outcomes; nevertheless, hospitalized patients with HF remain at considerable risk of functional decline, recurrent health-care utilization, and substantial resource consumption (2, 3). Early postdischarge risk is determined not only by cardiac pump function but also by comorbidities, nutritional status, and the quality of transitional care. Therefore, identifying clinically meaningful subgroups with distinct biological characteristics and short-term risks using routinely available clinical data may facilitate more individualized inpatient management and allocation of health-care resources.

Heart failure is not a single disease entity. Its clinical manifestations arise from the interplay among underlying etiology, cardiac structural and functional abnormalities, cardiorenal interactions, hematologic alterations, and inflammatory–nutritional status. Left ventricular ejection fraction remains a major component of contemporary HF classification and therapeutic decision-making; however, LVEF is a continuous and dynamic measure, and a single cutoff cannot adequately characterize differences in cardiorenal function, anemia, inflammatory burden, and systemic vulnerability (4). Even among patients within the same LVEF category, symptoms, functional status, and clinical outcomes may vary substantially. Accordingly, conventional LVEF-based stratification may fail to capture multidimensional patterns jointly reflected by routinely measured laboratory variables.

Nutritional status represents an important but frequently underrecognized component of HF heterogeneity. Appetite and dietary intake, gastrointestinal absorption, venous congestion, chronic inflammation, comorbidities, and therapeutic interventions may all influence nutritional status, whereas nutritional impairment may reduce treatment tolerance and functional recovery (5). Recent studies have demonstrated that malnutrition is common among hospitalized patients with HF and is associated with mortality and prolonged hospitalization; however, considerable heterogeneity exists across studies in both assessment methods and patient populations (6). Multidimensional nutritional assessment may provide more clinically relevant information than body weight or isolated laboratory measurements alone (7). Comprehensive assessment of patients with HF therefore requires consideration of dietary intake, laboratory biomarkers, inflammatory status, and overall clinical condition (8, 9).

Routine laboratory measurements provide an inexpensive, reproducible, and widely accessible approach to nutritional risk characterization. PNI integrates serum albumin and lymphocyte count, whereas ALI incorporates BMI, albumin, and the neutrophil-to-lymphocyte ratio to characterize inflammatory–nutritional status. Lower PNI and ALI values have previously been associated with adverse outcomes in HF (10, 11). However, individual composite indices cannot fully capture the interplay among inflammation, anemia, renal dysfunction, metabolic abnormalities, and nutritional status. Integrating routinely available laboratory measurements through unsupervised phenotyping may help identify clinically distinct subgroups with different nutritional and functional risks and provide a pragmatic basis for nutritional screening and rehabilitation-related risk assessment.

The integration of data science with clinical medicine provides new opportunities for more refined phenotyping of HF. Previous studies have used electronic health records and unsupervised machine-learning approaches to identify subgroups characterized by distinct comorbidity profiles, biological features, and clinical outcomes (12, 13). Nevertheless, existing studies are often limited by restricted LVEF ranges or limited variable availability, and laboratory-based phenotyping across the full LVEF spectrum with explicit emphasis on nutritional and inflammatory domains remains incompletely characterized. Thirty-day same-center all-cause unplanned readmission reflects early postdischarge vulnerability and is also an important measure of health-care utilization in HF (14, 15). We therefore performed unsupervised clustering based on routine admission laboratory measurements to identify inflammatory–nutritional phenotypes in hospitalized patients with HF and examined their associations with LVEF, severe functional limitation, and 30-day same-center all-cause unplanned readmission, while further evaluating the relationships of ALI and PNI with clinical outcomes.

2 Methods

2.1 Study design and study population

This was a single-center retrospective cohort study of patients hospitalized with HF at Beijing Anzhen Hospital, Capital Medical University, between June 2024 and December 2025, whose principal discharge diagnosis was HF as assessed by clinicians. For patients with multiple eligible records, only one qualifying hospitalization was retained.

The study collected biochemical panel testing; some patients underwent only individual biochemical tests and therefore had grouped missingness across multiple clustering variables. The primary clustering analysis included only patients with complete data for all 15 variables. Among 1,588 eligible patients, 1,455 had complete data for all 15 candidate clustering variables and were included in the primary complete-case clustering analysis. The remaining 133 patients were retained for full-cohort descriptions and missing-data sensitivity analyses. The detailed inclusion and exclusion process is shown in Figure 1.

Study population and analytic cohorts. ACS, acute coronary syndrome; ARI, adjusted Rand index; LVEF, left ventricular ejection fraction; MI, myocardial infarction; NMI, normalized mutual information; NYHA, New York Heart Association.
Figure 1. Study population and analytic cohorts. ACS, acute coronary syndrome; ARI, adjusted Rand index; LVEF, left ventricular ejection fraction; MI, myocardial infarction; NMI, normalized mutual information; NYHA, New York Heart Association.

The study was approved by the Ethics Committee of Beijing Anzhen Hospital, Capital Medical University (approval No. 2026099x). Given the retrospective use of deidentified clinical data, the requirement for informed consent was waived by the ethics committee.

2.2 Data sources and variable definitions

Demographic characteristics, comorbidities, cardiac functional measures, and laboratory data were obtained from the hospital information system and laboratory information system. Basic clinical variables included age, sex, BMI, hypertension, diabetes mellitus, and dyslipidemia. Valid sample sizes and missingness for all 78 source variables are provided in Supplementary Table S1.

The indicators used in this study were obtained within 48 h after admission and are usually completed before the treatment of heart failure such as major venous diuresis. The first available values of each indicator were taken within this time window. Detailed formulas and units for the derived indices are provided in Supplementary Table S2.

2.3 Clinical anchors and clinical outcome

NYHA class was obtained from the initial admission assessment and determined by the attending physician before clinical stabilization or adequate diuresis. Severe functional limitation was defined as NYHA functional class III or IV, whereas NYHA class I or II served as the reference category.

Thirty-day same-center all-cause unplanned readmission was defined as the first unplanned rehospitalization for any cause occurring within 30 days after discharge from the index hospitalization. Planned admissions, including prescheduled elective surgery or interventional procedures, device implantation, scheduled diagnostic evaluations or treatments, and other prearranged hospitalizations, were not counted as outcome events. Planned and unplanned readmissions were manually adjudicated by the study investigators through medical-record review. Readmission outcomes were identified only from the Beijing Anzhen Hospital information system; readmissions to other hospitals were unavailable. In-hospital deaths were excluded. All included patients had a follow-up record at this hospital more than 30 days after discharge, and no deaths within 30 days after discharge were identified. For patients with multiple qualifying unplanned readmissions within 30 days, only the first event was counted. Patients without a qualifying readmission within 30 days were classified as having no event.

Continuous LVEF, LVEF <50%, and severe functional limitation were used as post-clustering clinical anchors, whereas 30-day same-center all-cause unplanned readmission was evaluated as the short-term clinical outcome. None of these variables were included in the primary clustering model.

2.4 Candidate clustering variables and data preprocessing

Fifteen admission variables were prespecified for the primary clustering analysis: BMI, albumin, hemoglobin, red blood cell distribution width–coefficient of variation (RDW-CV), lymphocyte count, neutrophil count, monocyte count, high-sensitivity C-reactive protein (hs-CRP), platelet count, total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), glucose, estimated glomerular filtration rate (eGFR), and uric acid.

Variables were selected to provide clinically meaningful coverage of anthropometric, nutritional, hematologic, inflammatory, lipid metabolic, glucose metabolic, and renal domains while minimizing substantial redundancy. The Spearman correlation structure among candidate variables is shown in Supplementary Figure S1.

Nonnegative variables with a raw skewness >1 were transformed using log(1 + x); all other variables were retained on their original scales. All 15 variables were subsequently standardized to z scores with a mean of 0 and a standard deviation of 1. The primary clustering analysis was performed in the 1,455 patients with complete data for all candidate variables.

2.5 Unsupervised clustering, model selection, and phenogroup labeling

K-means clustering, Gaussian mixture modeling, and DBSCAN were compared using the same standardized 15-variable matrix.

For K-means clustering, solutions with K = 2 through K = 6 were evaluated using Lloyd’s algorithm, Euclidean distance, 100 random initializations, a maximum of 1,000 iterations, and a fixed random seed of 20,260,830. For each solution, the silhouette coefficient, Calinski-Harabasz index, Davies-Bouldin index, and cluster sizes were recorded.

For GMMs, solutions with K = 2 through K = 6 were evaluated under full, tied, diagonal, and spherical covariance structures with 10 initializations. Bayesian information criterion (BIC) and internal clustering metrics were recorded. DBSCAN was screened across 40 prespecified combinations of minimum-sample requirements and empirical k-distance quantiles. Complete parameter-screening results are presented in Supplementary Table S3.

The final clustering solution was selected by jointly considering internal separation, resampling stability, cluster size, and clinical interpretability rather than relying on a single internal metric; outcome variables were not used for solution selection, and K = 3 was treated only as the primary exploratory solution. Principal component analysis (PCA) was used only for two-dimensional visualization. Phenogroups were descriptively labeled according to the relative standardized levels of the 15 clustering variables within each group.

2.6 Cluster stability and sensitivity analyses

Cluster stability was evaluated using 500 bootstrap resamples. For each resample, candidate clustering models were refitted at the patient level, and cluster labels were aligned with the reference solution using one-to-one matching that maximized Jaccard overlap. Mean cluster-specific and overall matched Jaccard coefficients with 95% confidence intervals (CIs) were reported. K-means solutions with K = 2 through K = 5 and tied-covariance GMM solutions with K = 3 through K = 5 were evaluated.

Missing-data sensitivity analyses included all 1,588 patients and used 30 multiple imputations and median imputation. NYHA class and the readmission outcome were excluded from the imputation process. K-means clustering with K = 3 was refitted and labels were aligned in each imputed data set, and effect estimates from multiple imputation were pooled using Rubin’s rules. A K = 2 sensitivity analysis using the same preprocessing procedures and parameters was also performed in the 1,455 complete cases (Supplementary Tables S5, S6).

2.7 Statistical analysis

All statistical analyses were performed using Python 3.12.3, primarily with pandas 2.3.3, NumPy 1.26.4, SciPy 1.16.3, scikit-learn 1.8.0, and statsmodels 0.14.6. Continuous variables are presented as mean ± standard deviation or median (25th–75th percentile), as appropriate, and categorical variables as counts and percentages. Comparisons between LVEF categories were performed using Welch’s t test or the Mann–Whitney U test for continuous variables and the chi-square or Fisher exact test for categorical variables. Comparisons across the three phenogroups were performed using the Kruskal-Wallis test for continuous variables and the chi-square test for categorical variables.

Differences in inflammatory–nutritional indices across phenogroups were descriptively evaluated using pairwise areas under the receiver operating characteristic curve (AUCs). Because several composite indices shared constituent variables with the clustering inputs, these AUCs were used only to characterize phenotypic differences and were not interpreted as independent validation or predictive performance.

Using Cluster 3 as the reference group, models for continuous LVEF and LVEF <50% were adjusted for age, sex, hypertension, diabetes mellitus, and dyslipidemia. Associations of phenogroup membership with severe functional limitation and 30-day same-center all-cause unplanned readmission were evaluated using logistic regression with HC3 robust standard errors. The severe functional limitation model was adjusted for age, sex, hypertension, diabetes mellitus, dyslipidemia, continuous LVEF, and log-transformed NT-proBNP; the readmission model was additionally adjusted for NYHA class III–IV.

ALI and PNI quartile and restricted cubic spline models used the same adjustment variables. Age-, sex-, and LVEF-stratified analyses were exploratory and were not adjusted for multiple testing. Effect estimates are reported with 95% CIs. All tests were two-sided, and p < 0.05 was considered the reference threshold for statistical significance.

3 Results

3.1 Study population and baseline characteristics

A total of 1,588 adult patients hospitalized with HF were included. Of these, 1,332 (83.9%) had LVEF <50% and 256 (16.1%) had LVEF ≥50%. The mean age was 60.03 ± 14.05 years, and 1,180 patients (74.3%) were men.

Severe functional limitation was present in 379 patients (23.9%), and 214 patients (13.5%) experienced 30-day same-center all-cause unplanned readmission. Severe functional limitation was more common among patients with LVEF ≥50% than among those with LVEF <50% (34.0% vs. 21.9%, p < 0.001), whereas 30-day same-center all-cause unplanned readmission rates were similar between the two LVEF categories (12.9% vs. 13.6%, p = 0.765). Additional clinical, laboratory, and inflammatory–nutritional characteristics are presented in Table 1.

Table 1. Baseline characteristics according to LVEF category.

CharacteristicSummaryOverall (N = 1,588)LVEF <50%LVEF > = 50%Valid n (<50%)Valid n (> = 50%)p value
Age, yearsMean ± SD60.03 ± 14.0559.99 ± 14.3860.24 ± 12.251,3322560.771
Male sexn (%)1,180 (74.3%)983 (73.8%)197 (77.0%)1,3322560.327
BMI, kg/m2Mean ± SD23.60 ± 3.1523.52 ± 3.1424.00 ± 3.141,3312560.025
Hypertensionn (%)886 (55.8%)732 (55.0%)154 (60.2%)1,3322560.143
Diabetes mellitusn (%)612 (38.5%)507 (38.1%)105 (41.0%)1,3322560.413
Dyslipidemian (%)693 (43.6%)573 (43.0%)120 (46.9%)1,3322560.284
LVEF, %Mean ± SD35.73 ± 13.7231.00 ± 8.8460.32 ± 6.351,332256<0.001
Severe functional limitationn (%)379 (23.9%)292 (21.9%)87 (34.0%)1,332256<0.001
30-day same-center all-cause unplanned readmissionn (%)214 (13.5%)181 (13.6%)33 (12.9%)1,3322560.765
Hemoglobin, g/LMean ± SD135.82 ± 26.22137.01 ± 26.30129.49 ± 24.881,332251<0.001
RDW-CV, %Median [IQR]13.50 [12.90, 14.80]13.60 [12.90, 14.80]13.40 [12.90, 14.80]1,3322510.497
Lymphocyte, 10^9/LMedian [IQR]1.46 [1.03, 1.92]1.48 [1.05, 1.93]1.36 [0.89, 1.85]1,3322510.013
Neutrophil, 10^9/LMedian [IQR]4.56 [3.48, 6.08]4.52 [3.48, 5.99]4.68 [3.52, 6.68]1,3322510.042
Monocyte, 10^9/LMedian [IQR]0.44 [0.34, 0.57]0.43 [0.34, 0.56]0.47 [0.35, 0.61]1,3322510.027
Platelet, 10^9/LMedian [IQR]188.00 [149.00, 233.00]188.00 [150.00, 233.00]190.00 [147.00, 232.00]1,3322510.989
Albumin, g/LMean ± SD43.06 ± 4.2743.16 ± 4.2542.00 ± 4.431,3301290.005
hs-CRP, mg/LMedian [IQR]1.50 [0.68, 4.38]1.48 [0.65, 4.25]2.20 [0.83, 7.97]1,3311290.007
Total cholesterol, mmol/LMedian [IQR]3.70 [3.08, 4.55]3.69 [3.08, 4.55]3.80 [3.00, 4.79]1,3301280.471
Triglycerides, mmol/LMedian [IQR]1.33 [0.99, 1.87]1.34 [0.99, 1.87]1.29 [0.99, 1.92]1,3301280.959
HDL-C, mmol/LMean ± SD1.11 ± 0.291.12 ± 0.291.05 ± 0.301,3301280.027
Glucose, mmol/LMedian [IQR]5.61 [4.94, 6.81]5.63 [4.97, 6.83]5.33 [4.74, 6.33]1,3311290.02
eGFR, mL/min/1.73 m2Mean ± SD77.42 ± 24.0677.61 ± 23.9575.46 ± 25.201,3301280.355
Uric acid, μmol/LMedian [IQR]337.30 [270.10, 423.75]336.20 [269.25, 420.15]357.15 [271.65, 493.42]1,3311280.032
NT-proBNP, pg./mLMedian [IQR]1165.00 [423.75, 3181.79]1177.00 [438.75, 3138.99]1140.00 [359.64, 3587.03]1,3322560.755
PNIMedian [IQR]51.30 [47.55, 54.30]51.40 [47.60, 54.30]49.85 [46.50, 53.30]1,3301290.042
CONUT scoreMean ± SD2.46 ± 1.562.45 ± 1.542.56 ± 1.751,3291280.489
HALPMedian [IQR]47.06 [29.92, 66.78]47.81 [30.98, 66.80]39.58 [18.92, 66.16]1,3301290.011
hsCARMedian [IQR]0.03 [0.02, 0.11]0.03 [0.01, 0.10]0.05 [0.02, 0.19]1,3301290.005
ALIMedian [IQR]33.41 [19.97, 50.95]33.41 [20.19, 50.33]33.43 [15.10, 57.17]1,3291290.836

ALI, advanced lung cancer inflammation index; BMI, body mass index; CALLY, C-reactive protein-albumin-lymphocyte score; CONUT, Controlling Nutritional Status; eGFR, estimated glomerular filtration rate; HALP, hemoglobin-albumin-lymphocyte-platelet; HDL-C, high-density lipoprotein cholesterol; hsCAR, high-sensitivity C-reactive protein-to-albumin ratio; hs-CRP, high-sensitivity C-reactive protein; IQR, interquartile range; LVEF, left ventricular ejection fraction; NPS, Naples prognostic score; NT-proBNP, N-terminal pro-B-type natriuretic peptide; PNI, prognostic nutritional index; RAR, red cell distribution width-to-albumin ratio; RDW-CV, red blood cell distribution width-coefficient of variation; SD, standard deviation.

3.2 Candidate variables, preprocessing, and selection of the clustering solution

Fifteen admission variables were prespecified for clustering, covering anthropometric, nutritional, hematologic, inflammatory, lipid metabolic, glucose metabolic, and renal domains. Missingness ranged from 0.1 to 8.2%. Nonnegative variables with raw skewness >1 were transformed using log(1 + x), after which all variables were standardized to z scores. A total of 1,455 patients had complete data for all 15 variables and were included in the primary complete-case analysis (Table 2). The Spearman correlation matrix of the candidate variables is shown in Supplementary Figure S1. The proportions of complete cases were 99.6% among patients with LVEF <50 and 50.0% among those with LVEF ≥50%, indicating differential complete-case selection (Supplementary Table S7).

Table 2. Candidate clustering variables, data preprocessing, and inclusion status.

VariableValid nMissing (%)Raw skewnessTransformation
BMI1,5870.10.431None
Albumin1,4598.1−0.757None
Hemoglobin1,5830.3−0.397None
RDW-CV1,5830.32.525log1p
Lymphocyte1,5830.318.511log1p
Neutrophil1,5830.32.737log1p
Monocyte1,5830.32.845log1p
hs-CRP1,4608.17.608log1p
Platelet1,5830.31.552log1p
TC1,4588.21.162log1p
TG1,4588.210.516log1p
HDL-C1,4588.20.913None
Glucose1,4608.12.479log1p
eGFR1,4588.2−0.417None
Uric acid1,4598.11.244log1p

BMI, body mass index; eGFR, estimated glomerular filtration rate; HDL-C, high-density lipoprotein cholesterol; hs-CRP, high-sensitivity C-reactive protein; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro-B-type natriuretic peptide; RDW-CV, red blood cell distribution width-coefficient of variation; TC, total cholesterol; TG, triglycerides.

The clustering-method comparison is shown in Figure 2, with complete parameter-screening results provided in Supplementary Table S3. For K-means with K = 3, the silhouette coefficient was 0.107, the mean bootstrap matched Jaccard coefficient was 0.823, and the mean Jaccard coefficient of the least stable cluster was 0.784 (Supplementary Table S4). The lowest BIC for GMM occurred at K = 5; however, the smallest subgroup comprised only 3.64% of participants. DBSCAN failed to identify a valid multi-cluster solution across all 40 parameter settings.

Comparison of clustering methods and selection of an exploratory K = 3 solution. **(A)** Shows the silhouette coefficient for different numbers of clusters; **(B)** Shows the Davies-Bouldin index; **(C)** Shows the matched Jaccard coefficient and 95% CI obtained from 500 bootstrap repetitions; **(D)** Shows the BIC for GMMs with different covariance structures; **(E)** Shows the results of DBSCAN parameter screening; **(F)** Summarizes the final selection. Based on a comprehensive evaluation of internal separation, cluster size, stability, and interpretability, K-means with K = 3 was selected as the primary exploratory solution. BIC, Bayesian information criterion; CI, confidence interval; DBSCAN, density-based spatial clustering of applications with noise; GMM, Gaussian mixture model; K, number of clusters.
Figure 2. Comparison of clustering methods and selection of an exploratory K = 3 solution. (A) Shows the silhouette coefficient for different numbers of clusters; (B) Shows the Davies-Bouldin index; (C) Shows the matched Jaccard coefficient and 95% CI obtained from 500 bootstrap repetitions; (D) Shows the BIC for GMMs with different covariance structures; (E) Shows the results of DBSCAN parameter screening; (F) Summarizes the final selection. Based on a comprehensive evaluation of internal separation, cluster size, stability, and interpretability, K-means with K = 3 was selected as the primary exploratory solution. BIC, Bayesian information criterion; CI, confidence interval; DBSCAN, density-based spatial clustering of applications with noise; GMM, Gaussian mixture model; K, number of clusters.

Considering internal separation, stability, subgroup size, and clinical interpretability, K-means clustering with K = 3 was selected as the primary exploratory solution. After multiple imputation and median imputation, agreement of the K = 3 labels with the primary analysis was 97.5 and 97.7%, respectively, and the directions of association with both outcomes were consistent (Supplementary Table S5). The K = 2 solution had a silhouette coefficient of 0.126, higher than the 0.107 observed for K = 3, but the directions of association with both outcomes were consistent (Supplementary Table S6). K = 3 was therefore retained as the clinically interpretable primary exploratory solution.

3.3 Identification and characteristics of the three exploratory HF phenogroups

In the two-dimensional PCA projection, principal components 1 and 2 explained 14.7 and 12.0% of the total variance, respectively. The three phenogroups showed different relative distributions in the two-dimensional space, although substantial overlap remained. Therefore, the PCA plot was used for visualization rather than to define deterministic cluster boundaries.

K-means clustering with K = 3 identified three exploratory HF phenogroups. Cluster 1, comprising 425 patients (29.2%), was characterized as an inflammatory–cardiorenal–nutritional vulnerability phenotype. This group was characterized by higher hs-CRP, glucose, and uric acid levels and lower albumin, eGFR, and HDL-C levels.

Cluster 2 included 264 patients (18.1%) and was characterized as an anemia–immune dysregulation phenotype, with lower hemoglobin, lymphocyte count, and platelet count and higher RDW-CV, neutrophil count, and monocyte count (Figure 3).

Principal component analysis visualization of the three exploratory phenogroups. The scatter plots are based on the primary complete-case cohort (*n* = 1,455). PC1 and PC2 explain 14.7 and 12.0% of the variance, respectively. PCA, principal component analysis; PC1, first principal component; PC2, second principal component.
Figure 3. Principal component analysis visualization of the three exploratory phenogroups. The scatter plots are based on the primary complete-case cohort (n = 1,455). PC1 and PC2 explain 14.7 and 12.0% of the variance, respectively. PCA, principal component analysis; PC1, first principal component; PC2, second principal component.

Cluster 3 comprised 766 patients (52.6%) and represented a relatively preserved laboratory phenotype.

Sex distribution, BMI, and the prevalence of hypertension, diabetes mellitus, and dyslipidemia were similar across phenogroups. In contrast, age, severe functional limitation, 30-day same-center all-cause unplanned readmission, and most laboratory and inflammatory–nutritional indices differed significantly across groups (Table 3).

Table 3. Clinical, laboratory, and 30-day same-center all-cause unplanned readmission characteristics across the three exploratory phenogroups.

CharacteristicSummaryOverall (N = 1,455)Cluster 1 (n = 425)Cluster 2 (n = 264)Cluster 3 (n = 766)P value
Age, yearsMean ± SD60.00 ± 14.2061.44 ± 14.2160.33 ± 12.6259.09 ± 14.650.016
Male sexn (%)1,076 (74.0%)309 (72.7%)199 (75.4%)568 (74.2%)0.727
BMI, kg/m2Mean ± SD23.55 ± 3.1323.40 ± 3.0923.43 ± 3.1623.68 ± 3.140.183
Hypertensionn (%)807 (55.5%)233 (54.8%)154 (58.3%)420 (54.8%)0.584
Diabetes mellitusn (%)558 (38.4%)167 (39.3%)108 (40.9%)283 (36.9%)0.465
Dyslipidemian (%)631 (43.4%)186 (43.8%)114 (43.2%)331 (43.2%)0.981
LVEF, %Mean ± SD33.54 ± 11.9433.98 ± 12.6333.73 ± 12.1033.23 ± 11.480.954
Severe functional limitationn (%)371 (25.5%)155 (36.5%)119 (45.1%)97 (12.7%)<0.001
30-day same-center all-cause unplanned readmissionn (%)196 (13.5%)71 (16.7%)55 (20.8%)70 (9.1%)<0.001
Hemoglobin, g/LMean ± SD136.56 ± 26.33139.71 ± 23.28104.67 ± 21.79145.81 ± 20.30<0.001
RDW-CV, %Median [IQR]13.60 [13.00, 14.90]13.50 [13.00, 14.80]15.90 [14.60, 17.12]13.30 [12.80, 13.90]<0.001
Lymphocyte, 10^9/LMedian [IQR]1.47 [1.05, 1.93]1.53 [1.13, 1.89]0.81 [0.52, 1.14]1.69 [1.27, 2.13]<0.001
Neutrophil, 10^9/LMedian [IQR]4.53 [3.48, 5.99]4.31 [3.46, 5.43]7.47 [4.97, 10.60]4.19 [3.29, 5.36]<0.001
Monocyte, 10^9/LMedian [IQR]0.43 [0.34, 0.56]0.44 [0.35, 0.53]0.53 [0.40, 0.71]0.41 [0.33, 0.53]<0.001
Platelet, 10^9/LMedian [IQR]189.00 [150.00, 234.00]193.00 [155.00, 239.00]147.50 [95.75, 205.75]195.00 [160.25, 237.00]<0.001
Albumin, g/LMean ± SD43.06 ± 4.2840.05 ± 4.3643.49 ± 3.9344.58 ± 3.39<0.001
hs-CRP, mg/LMedian [IQR]1.50 [0.68, 4.39]5.45 [2.38, 14.70]1.40 [0.69, 3.64]0.92 [0.50, 1.85]<0.001
Total cholesterol, mmol/LMedian [IQR]3.69 [3.08, 4.55]3.62 [3.00, 4.37]3.83 [3.03, 4.84]3.69 [3.12, 4.55]0.172
Triglycerides, mmol/LMedian [IQR]1.33 [0.99, 1.87]1.43 [1.07, 2.10]1.40 [1.02, 1.96]1.23 [0.95, 1.73]<0.001
HDL-C, mmol/LMean ± SD1.11 ± 0.290.96 ± 0.241.08 ± 0.271.20 ± 0.29<0.001
Glucose, mmol/LMedian [IQR]5.60 [4.94, 6.81]6.02 [5.11, 8.09]5.54 [4.85, 6.68]5.49 [4.90, 6.33]<0.001
eGFR, mL/min/1.73 m2Mean ± SD77.41 ± 24.0461.60 ± 24.0978.27 ± 24.0285.89 ± 19.17<0.001
Uric acid, μmol/LMedian [IQR]337.30 [270.00, 423.75]431.60 [351.60, 537.10]343.80 [273.25, 428.15]298.50 [243.80, 356.23]<0.001
NT-proBNP, pg./mLMedian [IQR]1192.15 [443.77, 3236.64]1337.00 [533.00, 3448.00]1254.00 [550.99, 4395.75]1080.52 [387.25, 2829.25]0.002
PNIMedian [IQR]51.30 [47.55, 54.30]48.55 [44.60, 52.00]48.25 [45.08, 51.30]53.10 [50.50, 56.14]<0.001
CONUT scoreMean ± SD2.46 ± 1.562.62 ± 1.693.50 ± 1.492.02 ± 1.30<0.001
HALPMedian [IQR]47.06 [29.92, 66.78]44.52 [29.59, 61.62]23.67 [13.59, 37.68]55.03 [40.26, 75.96]<0.001
hsCARMedian [IQR]0.03 [0.02, 0.11]0.14 [0.06, 0.39]0.03 [0.02, 0.09]0.02 [0.01, 0.04]<0.001
ALIMedian [IQR]33.42 [19.98, 50.93]33.36 [20.53, 47.05]11.73 [5.65, 21.08]41.45 [28.96, 58.43]<0.001
CALLYMedian [IQR]38.18 [12.29, 98.74]11.35 [3.57, 28.37]22.66 [6.83, 53.60]77.93 [34.50, 152.40]<0.001
NPSMean ± SD2.16 ± 1.062.41 ± 1.102.76 ± 0.751.81 ± 1.00<0.001
RARMedian [IQR]0.32 [0.29, 0.36]0.34 [0.31, 0.38]0.37 [0.33, 0.41]0.30 [0.28, 0.32]<0.001

Three exploratory phenogroups were identified through K-means clustering of 1,455 complete cases (K = 3, random seed 20,260,830, n_init = 100). Cluster 1 represents the inflammatory–cardiorenal–nutritional vulnerability phenotype; Cluster 2 represents the anemia–immune dysregulation phenotype; and Cluster 3 represents the relatively preserved laboratory phenotype. ALI, advanced lung cancer inflammation index; BMI, body mass index; CALLY, C-reactive protein-albumin-lymphocyte score; CONUT, Controlling Nutritional Status; eGFR, estimated glomerular filtration rate; HALP, hemoglobin-albumin-lymphocyte-platelet; HDL-C, high-density lipoprotein cholesterol; hsCAR, high-sensitivity C-reactive protein-to-albumin ratio; hs-CRP, high-sensitivity C-reactive protein; IQR, interquartile range; LVEF, left ventricular ejection fraction; NPS, Naples prognostic score; NT-proBNP, N-terminal pro-B-type natriuretic peptide; PNI, prognostic nutritional index; RAR, red cell distribution width-to-albumin ratio; RDW-CV, red blood cell distribution width-coefficient of variation; SD, standard deviation.

3.4 Adjusted associations of phenogroups with clinical anchors and clinical outcome

Using Cluster 3 as the reference group and adjusting for age, sex, hypertension, diabetes mellitus, and dyslipidemia, no significant differences in continuous LVEF were observed for Cluster 1 or Cluster 2.

Cluster 1 was associated with lower odds of LVEF <50% (OR, 0.574; 95% CI, 0.380–0.867; p = 0.008), whereas the corresponding association for Cluster 2 did not reach conventional statistical significance (OR, 0.623; 95% CI, 0.385–1.008; p = 0.054).

After enhanced adjustment, compared with Cluster 3, the ORs for severe functional limitation were 3.814 (95% CI, 2.813–5.170) for Cluster 1 and 5.594 (95% CI, 3.928–7.965) for Cluster 2. The corresponding ORs for 30-day same-center all-cause unplanned readmission were 1.966 (95% CI, 1.364–2.835) and 2.616 (95% CI, 1.740–3.932), respectively (Table 4).

Table 4. Adjusted associations of exploratory phenogroups with clinical anchors and 30-day same-center all-cause unplanned readmission.

OutcomeComparisonAdjusted effect (95% CI)P value
LVEF (%)Cluster 1 vs. Cluster 30.728 (−0.732 to 2.189)0.328
LVEF (%)Cluster 2 vs. Cluster 30.413 (−1.250 to 2.077)0.626
LVEF <50%Cluster 1 vs. Cluster 30.574 (0.380 to 0.867)0.008
LVEF <50%Cluster 2 vs. Cluster 30.623 (0.385 to 1.008)0.054
Severe functional limitationCluster 1 vs. Cluster 33.814 (2.813 to 5.170)<0.001
Severe functional limitationCluster 2 vs. Cluster 35.594 (3.928 to 7.965)<0.001
30-day same-center all-cause unplanned readmissionCluster 1 vs. Cluster 31.966 (1.364 to 2.835)<0.001
30-day same-center all-cause unplanned readmissionCluster 2 vs. Cluster 32.616 (1.740 to 3.932)<0.001

Cluster 3 served as the reference group. Models for LVEF-related outcomes were adjusted for age, sex, hypertension, diabetes mellitus, and dyslipidemia. The severe functional limitation model used logistic regression with HC3 robust standard errors and was adjusted for age, sex, hypertension, diabetes mellitus, dyslipidemia, continuous LVEF, and log-transformed NT-proBNP; the readmission model was additionally adjusted for NYHA class III–IV. CI, confidence interval; LVEF, left ventricular ejection fraction; OR, odds ratio.

3.5 Distribution and descriptive discrimination of inflammatory–nutritional indices across phenogroups

Among the 11 inflammatory–nutritional indices evaluated, ALI showed marked descriptive differences across phenogroups, with the highest mean pairwise AUC of 0.790 (95% CI, 0.773–0.808) and a minimum pairwise AUC of 0.625 (95% CI, 0.590–0.661). For the traditional PNI, the three-group effect size was 0.236 (95% CI, 0.203–0.278), and the mean pairwise AUC was 0.696 (95% CI, 0.680–0.715).

Because ALI, PNI, and several other composite indices share constituent variables with the clustering inputs, these findings were interpreted only as descriptive characterization of inflammatory–nutritional differences across phenogroups and not as independent phenotype validation or predictive performance (Table 5). Formulas and units for all inflammatory–nutritional indices are provided in Supplementary Table S2.

Table 5. Distribution and descriptive pairwise discrimination of inflammatory–nutritional indices across the exploratory phenogroups.

IndexCluster medians (C1/C2/C3)Epsilon-squared (95% CI)Mean pairwise AUC (95% CI)
ALI33.36/11.73/41.450.289 (0.256 to 0.323)0.790 (0.773 to 0.808)
PNI48.55/48.25/53.100.236 (0.203 to 0.278)0.696 (0.680 to 0.715)
CALLY11.35/22.66/77.930.324 (0.285 to 0.366)0.746 (0.728 to 0.767)
RAR0.34/0.37/0.300.275 (0.233 to 0.314)0.740 (0.720 to 0.759)
HALP44.52/23.67/55.030.199 (0.166 to 0.238)0.737 (0.716 to 0.756)
hsCAR0.14/0.03/0.020.298 (0.264 to 0.336)0.745 (0.724 to 0.764)
LCR0.29/0.52/1.740.310 (0.269 to 0.353)0.739 (0.720 to 0.761)
NAR0.11/0.17/0.090.150 (0.118 to 0.187)0.708 (0.687 to 0.727)
NPS2.00/3.00/2.000.131 (0.104 to 0.163)0.668 (0.649 to 0.686)
CONUT score2.00/4.00/2.000.127 (0.098 to 0.161)0.683 (0.664 to 0.702)
MAR0.01/0.01/0.010.060 (0.038 to 0.091)0.621 (0.596 to 0.643)

ε2 is based on the Kruskal-Wallis test; both the AUC and the confidence intervals for the effect size were estimated using 500 bootstrap repetitions. ALI, advanced lung cancer inflammation index; AUC, area under the receiver operating characteristic curve; CALLY, C-reactive protein-albumin-lymphocyte score; CI, confidence interval; CONUT, Controlling Nutritional Status; HALP, hemoglobin-albumin-lymphocyte-platelet; hsCAR, high-sensitivity C-reactive protein-to-albumin ratio; LCR, lymphocyte-to-C-reactive protein ratio; MAR, monocyte-to-albumin ratio; NAR, neutrophil-to-albumin ratio; NPS, Naples prognostic score; PNI, prognostic nutritional index; RAR, red cell distribution width-to-albumin ratio.

3.6 Associations of ALI and PNI quartiles with severe functional limitation and 30-day same-center all-cause unplanned readmission

Using the lowest quartile (Q1) as the reference group, after enhanced adjustment, progressively higher quartiles of ALI and PNI were associated with lower odds of severe functional limitation (all p < 0.001).

For 30-day same-center all-cause unplanned readmission, ALI quartiles Q2–Q4 were associated with lower odds than Q1, whereas for PNI only Q4 was associated with lower odds than Q1 (OR, 0.533; 95% CI, 0.335–0.849; p = 0.008) (Table 6).

Table 6. Associations of ALI and PNI quartiles with severe functional limitation and 30-day same-center all-cause unplanned readmission.

IndexOutcomeQuartileN (events)Model 1Model 2Model 3
OR (95% CI)P valueOR (95% CI)P valueOR (95% CI)P value
ALISevere functional limitationQ1364 (184)ReferenceReferenceReference
ALISevere functional limitationQ2364 (78)0.267 (0.193 to 0.369)<0.0010.267 (0.193 to 0.369)<0.0010.278 (0.195 to 0.395)<0.001
ALISevere functional limitationQ3363 (59)0.190 (0.134 to 0.268)<0.0010.191 (0.135 to 0.270)<0.0010.173 (0.119 to 0.251)<0.001
ALISevere functional limitationQ4364 (50)0.156 (0.108 to 0.224)<0.0010.159 (0.110 to 0.229)<0.0010.148 (0.099 to 0.220)<0.001
ALI30-day same-center all-cause unplanned readmissionQ1364 (72)ReferenceReferenceReference
ALI30-day same-center all-cause unplanned readmissionQ2364 (42)0.529 (0.350 to 0.799)0.0020.533 (0.352 to 0.806)0.0030.549 (0.354 to 0.851)0.007
ALI30-day same-center all-cause unplanned readmissionQ3363 (47)0.603 (0.404 to 0.900)0.0130.605 (0.406 to 0.903)0.0140.617 (0.407 to 0.937)0.023
ALI30-day same-center all-cause unplanned readmissionQ4364 (35)0.431 (0.280 to 0.666)<0.0010.435 (0.280 to 0.674)<0.0010.451 (0.284 to 0.715)<0.001
PNISevere functional limitationQ1369 (180)ReferenceReferenceReference
PNISevere functional limitationQ2363 (93)0.362 (0.265 to 0.494)<0.0010.365 (0.267 to 0.498)<0.0010.373 (0.266 to 0.525)<0.001
PNISevere functional limitationQ3359 (54)0.186 (0.130 to 0.265)<0.0010.188 (0.132 to 0.267)<0.0010.179 (0.123 to 0.261)<0.001
PNISevere functional limitationQ4364 (44)0.144 (0.099 to 0.210)<0.0010.149 (0.102 to 0.218)<0.0010.148 (0.099 to 0.222)<0.001
PNI30-day same-center all-cause unplanned readmissionQ1369 (64)ReferenceReferenceReference
PNI30-day same-center all-cause unplanned readmissionQ2363 (52)0.797 (0.535 to 1.187)0.2640.800 (0.537 to 1.193)0.2750.828 (0.552 to 1.240)0.359
PNI30-day same-center all-cause unplanned readmissionQ3359 (46)0.700 (0.465 to 1.056)0.0890.705 (0.467 to 1.065)0.0970.739 (0.482 to 1.135)0.167
PNI30-day same-center all-cause unplanned readmissionQ4364 (34)0.491 (0.315 to 0.766)0.0020.506 (0.324 to 0.790)0.0030.533 (0.335 to 0.849)0.008

Quartiles were defined as the 25th, 50th, and 75th percentiles in the complete-case cohort for the primary analysis (N = 1,455), with values equal to the cutoff points assigned to the lower quartile; the ALI cutoff points were 19.98, 33.42, and 50.93, and the PNI cutoff points were 47.55, 51.30, and 54.30, with Q1 as the reference. Model 1 was unadjusted; Model 2 was adjusted for age and sex; Model 3 was further adjusted for hypertension, diabetes mellitus, dyslipidemia, continuous LVEF, and log-transformed NT-proBNP, with NYHA class III–IV additionally included in the readmission model. Logistic regression used HC3 robust standard errors. All p values are two-sided. ALI, advanced lung cancer inflammation index; CI, confidence interval; OR, odds ratio; PNI, prognostic nutritional index.

3.7 Restricted cubic spline associations of inflammatory–nutritional indices with clinical outcomes

After enhanced adjustment, restricted cubic spline analyses demonstrated significant overall and nonlinear associations of both ALI and PNI with severe functional limitation (all overall and nonlinear p < 0.001) (Figure 4).

Restricted cubic spline associations of ALI and PNI with severe functional limitation and 30-day same-center all-cause unplanned readmission. Panels **(A,B)** show the associations of ALI and PNI, respectively, with severe functional limitation; panels **(C,D)** show their associations with 30-day same-center all-cause unplanned readmission. Solid lines represent adjusted odds ratios (ORs), and shaded areas indicate 95% confidence intervals (CIs); horizontal dashed lines indicate OR = 1, and vertical dashed lines indicate the median reference value for the corresponding index. Models used logistic regression with HC3 robust standard errors. The severe functional limitation models were adjusted for age, sex, hypertension, diabetes mellitus, dyslipidemia, continuous LVEF, and log-transformed NT-proBNP; the readmission models were additionally adjusted for NYHA class III–IV. Four knots were placed at P5, P35, P65, and P95. ALI, advanced lung cancer inflammation index; CI, confidence interval; OR, odds ratio; PNI, prognostic nutritional index; RCS, restricted cubic spline.
Figure 4. Restricted cubic spline associations of ALI and PNI with severe functional limitation and 30-day same-center all-cause unplanned readmission. Panels (A,B) show the associations of ALI and PNI, respectively, with severe functional limitation; panels (C,D) show their associations with 30-day same-center all-cause unplanned readmission. Solid lines represent adjusted odds ratios (ORs), and shaded areas indicate 95% confidence intervals (CIs); horizontal dashed lines indicate OR = 1, and vertical dashed lines indicate the median reference value for the corresponding index. Models used logistic regression with HC3 robust standard errors. The severe functional limitation models were adjusted for age, sex, hypertension, diabetes mellitus, dyslipidemia, continuous LVEF, and log-transformed NT-proBNP; the readmission models were additionally adjusted for NYHA class III–IV. Four knots were placed at P5, P35, P65, and P95. ALI, advanced lung cancer inflammation index; CI, confidence interval; OR, odds ratio; PNI, prognostic nutritional index; RCS, restricted cubic spline.

For 30-day same-center all-cause unplanned readmission, the overall and nonlinear p values were 0.011 and 0.133 for ALI and 0.046 and 0.115 for PNI, respectively.

Age-, sex-, and LVEF-stratified results were not adjusted for multiple testing and were considered hypothesis-generating only (Supplementary Figure S2).

4 Discussion

In this study, three exploratory phenogroups were identified among 1,455 hospitalized patients with HF using routine admission laboratory measurements. After enhanced adjustment, compared with the relatively preserved laboratory phenotype, the inflammatory–cardiorenal–nutritional vulnerability phenotype had ORs of 3.814 for severe functional limitation and 1.966 for 30-day same-center all-cause unplanned readmission, whereas the corresponding ORs for the anemia–immune dysregulation phenotype were 5.594 and 2.616. LVEF was similar across the three phenogroups, and after further adjustment for continuous LVEF and NT-proBNP, both vulnerable phenogroups remained associated with both outcomes, suggesting that phenogroup membership provided complementary association information beyond LVEF. After enhanced adjustment, ALI was associated with both outcomes, whereas the association of PNI with readmission was mainly observed in the highest quartile. These exploratory findings support the need for further comprehensive assessment.

The enhanced models and full-cohort sensitivity analyses yielded directionally consistent associations after adjustment for continuous LVEF and NT-proBNP. These findings indicate that routine laboratory measurements may capture complementary dimensions of systemic vulnerability beyond LVEF, but they do not establish incremental predictive performance or clinical utility.

Our findings are consistent with previous HF phenotyping studies demonstrating that data-driven subgroups based on comorbidities and laboratory variables may provide clinically relevant information beyond conventional LVEF classification (16). In HF with preserved ejection fraction, acute HF, and critically ill cardiac populations, phenogroups characterized by inflammation, cardiorenal dysfunction, anemia, and iron deficiency have repeatedly been identified (17–19). Similar multidimensional heterogeneity has been demonstrated using different clustering approaches and therapeutic cohorts (20, 21). The present study integrated these features into three exploratory phenogroups and described their associations with functional limitation and 30-day same-center all-cause unplanned readmission within a single framework. The low silhouette coefficient and overlap in the PCA projection indicate that these phenogroups should be interpreted as data-driven partitions within a continuum of clinical heterogeneity rather than as clearly separated biological entities; the K = 2 sensitivity analysis yielded directionally consistent outcome associations (Supplementary Table S6).

The anemia–immune dysregulation phenotype was characterized by lower hemoglobin, lymphocyte count, and platelet count, together with higher RDW-CV, neutrophil count, and monocyte count. Anemia is strongly associated with adverse clinical outcomes in HF (22), and RDW may reflect the combined effects of inflammation, nutritional status, renal dysfunction, and impaired erythropoiesis (23). Anemia may also reduce tissue oxygen delivery and exercise tolerance (24). Alterations in neutrophil, monocyte, and lymphocyte profiles reflect persistent inflammatory and immune activation and have been associated with adverse cardiovascular outcomes (25, 26). Collectively, these abnormalities characterize a state of anemia, inflammation, and immune dysregulation that may help explain the greater functional limitation and higher early readmission rate observed in this phenogroup.

The inflammatory–cardiorenal–nutritional vulnerability phenotype was characterized by higher hs-CRP, uric acid, and glucose levels and lower albumin, eGFR, and HDL-C levels. Venous congestion, impaired renal perfusion, and metabolic abnormalities may interact to increase systemic disease burden (27). Low albumin may reflect impaired nutritional status, inflammatory activity, and altered volume status (28), whereas elevated uric acid may indicate greater metabolic and oxidative stress (29). The combined pattern of higher hs-CRP and lower albumin further highlights the close interplay between inflammation and nutritional vulnerability (30). However, albumin and BMI are also influenced by inflammation, volume status, and other clinical factors and cannot independently diagnose malnutrition. This phenogroup therefore reflects a clinically coherent pattern of inflammatory activation, cardiorenal dysfunction, metabolic burden, and nutritional vulnerability.

Lower PNI has previously been associated with adverse outcomes in HF (11), and recent studies have also reported associations between ALI and HF outcomes (31, 32). However, ALI and PNI reflect laboratory-derived inflammatory–nutritional vulnerability. After enhanced adjustment, both indices were associated with severe functional limitation; ALI was associated with 30-day same-center all-cause unplanned readmission, whereas the association for PNI was mainly observed in the highest quartile. Because their component variables overlap with the clustering inputs and this was an observational study, these results do not provide independent validation or evidence of incremental predictive value, nor do they establish independent biological pathways or causal effects. ALI and PNI cannot replace formal nutritional assessments such as GLIM, SGA, or MNA. Their inverse associations with severe functional limitation were stronger among patients aged <65 years, possibly because functional status in older patients is also influenced by frailty, multimorbidity, and sarcopenia; however, this subgroup analysis was exploratory and should not be used to define age-specific thresholds.

These findings may complement the description of laboratory vulnerability but are insufficient to support treatment selection, risk stratification, or individualized management. Early readmission in HF is influenced jointly by cardiac disease severity, renal dysfunction, anemia, inflammation, nutritional status, and other comorbidities (33, 34), several of which also affect exercise capacity, functional recovery, and rehabilitation tolerance. Frailty, dietary and inflammatory factors, precision nutrition, outpatient follow-up, and transitional care may further influence functional recovery and postdischarge healthcare utilization in patients with HF (35–39).

Several limitations should be acknowledged. The single-center design and unequal proportions of complete cases across LVEF categories may have introduced selection bias; although the full-cohort imputation analyses yielded directionally consistent findings, this bias cannot be excluded. In addition, the NYHA class in this study was derived from the initial assessment upon admission and may be influenced by anemia, pulmonary disease, obesity, frailty, and other non cardiac factors, and may change with stable condition and diuretic treatment during hospitalization. Therefore, NYHA class III–IV were only used as clinical anchors for severe functional limitation in this study and should not be interpreted as specific indicators of heart failure severity, nor as independent biological validation for phenogroups. Readmission ascertainment captured only events at Beijing Anzhen Hospital, and readmissions to other hospitals may have been missed. In-hospital deaths were excluded, and only patients with a follow-up record at this hospital more than 30 days after discharge were included. Although these records confirmed survival at 30 days among included patients, out-of-hospital events may still have been missed and survivor-selection bias may remain, potentially leading to underestimation of readmission rates and conservative association estimates. The readmission findings should therefore be interpreted as associations with same-center events rather than as the complete 30-day all-cause readmission risk. Kidney function and hemoglobin were used to construct the phenogroups and were not repeatedly adjusted for in the outcome models. Previous HF hospitalization, HF etiology, atrial fibrillation, and discharge medications were not fully incorporated, and residual confounding may therefore remain. Future studies should integrate more complete information on lifestyle factors, pharmacologic therapy, and postdischarge management to further validate these phenogroups and their clinical implications.

5 Conclusion

Three phenogroups identified from routine admission laboratory measurements revealed substantial heterogeneity in inflammatory, nutritional, hematologic, and cardiorenal status among hospitalized patients with HF. After adjustment for continuous LVEF and NT-proBNP, the two vulnerable phenogroups remained associated with severe functional limitation and 30-day same-center all-cause unplanned readmission, with directionally consistent associations in the full-cohort and K = 2 sensitivity analyses. ALI and PNI reflect laboratory-derived inflammatory–nutritional vulnerability but cannot replace formal nutritional assessment or establish incremental predictive value. These findings are exploratory and require validation in prospective multicenter cohorts.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

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.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnut.2026.1985953/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Natalia Świątoniowska-Lonc, 4th Military Hospital of Wroclaw, Poland

Reviewed by: Can Baba Arin, Dr. Siyami Ersek Chest Cardiovascular Surgery Training and Research Hospital, Türkiye

Yasutaka Imamura, Kugayama Heart Clinic, Japan

Statements

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Data availability

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

Ethics approval

The studies involving humans were approved by Ethics Committee of Beijing Anzhen Hospital. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements. Written informed consent was not obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article because the study was approved by the Ethics Committee of Beijing Anzhen Hospital, Capital Medical University (Approval No. 2026099x). Given the retrospective use of deidentified clinical data, the requirement for informed consent was waived by the ethics committee.

Author contributions

YZ: Formal analysis, Methodology, Visualization, Writing – original draft. XZ: Investigation, Software, Supervision, Validation, Writing – review & editing. HS: Investigation, Methodology, Resources, Writing – review & editing. QH: Formal analysis, Investigation, Methodology, Writing – review & editing. JN: Conceptualization, Data curation, Funding acquisition, Supervision, Writing – review & editing.

References

  1. 1. Chen Q-F, Chen L, Katsouras CS, Liu C, Shi J, Liang D. Global burden of heart failure and its underlying causes in 204 countries and territories, 1990-2021. Eur Heart J Qual Care Clin Outcomes. (2025) 11: 493 – 509. doi:, doi:10.1093/ehjqcco/qcae110
  2. 2. Parizad R, Hatwal J, Taban Sadeghi M, Brar A, Desai R, Batta A. Clinical advances in heart failure with preserved ejection fraction: a systematic review of therapeutic and mechanistic evidence. Vasc Health Risk Manag. (2026) 22: 578698. doi:, doi:10.2147/VHRM.S578698
  3. 3. Teimourizad A, Jafari A, Esmaeilzadeh F. Budget impact analyses for treatment of heart failure. A systematic review. Heart Fail Rev. (2024) 29: 785 – 97. doi:, doi:10.1007/s10741-024-10397-8
  4. 4. Dimond MG, Ibrahim NE, Fiuzat M, McMurray JJV, Lindenfeld J, Ahmad T. Left ventricular ejection fraction and the future of heart failure phenotyping. JACC Heart Fail. (2024) 12: 451 – 60. doi:, doi:10.1016/j.jchf.2023.11.005
  5. 5. Pérez-Rivera J-Á, Bonilla-Palomas JL, Diez-Villanueva P, Jiménez-Méndez C, Solla-Suárez P, López-Azor JC. Nutritional intervention in older outpatients with chronic heart failure and malnutrition or at risk of malnutrition: the BOCADOS-IC randomized trial. Rev Esp Cardiol (Engl Ed). (2026): S1885-5857(26)00158–1. [online ahead of print]. doi:, doi:10.1016/j.rec.2026.07.001
  6. 6. Knobloch I d S, Zucatti KP, de Carvalho BZO, da Costa-Pereira JP, Hartmann AS, Razzera EL. Prevalence of malnutrition and its association with outcomes in heart failure: a systematic review and meta-analysis. Nutrition. (2025) 140: 112913. doi:, doi:10.1016/j.nut.2025.112913
  7. 7. Osório AF, Ribeiro ÉCT, Parahiba SM, Forte GC, Clausell NO, Souza GC. Prognostic value of nutritional screening tools in hospitalized patients with decompensated heart failure: a systematic review and meta-analysis. Nutr Res. (2023) 120: 1 – 19. doi:, doi:10.1016/j.nutres.2023.09.009
  8. 8. Ilonze OJ, Parsly Read-Button L, Cogswell R, Hackman A, Breathett K, Saltzman E. Controversies and conundrums in cardiac cachexia: key questions about wasting in patients with HFrEF. JACC Heart Fail. (2024) 12: 1645 – 60. doi:, doi:10.1016/j.jchf.2024.03.003
  9. 9. Belqaid K, Irving GF, Waldréus N. Nutritional interventions in older, frail persons with heart failure-a systematic narrative review. JAR Life. (2024) 13: 99 – 107. doi:, doi:10.14283/jarlife.2024.15
  10. 10. Amin AM, Ghaly R, Elbenawi H, Ewis A, Khan U, Elshaer KSM. Impact of advanced lung cancer inflammation index on all-cause mortality among patients with heart failure: a systematic review and meta-analysis with reconstructed time-to-event data. Cardiooncology. (2025) 11: 9. doi:, doi:10.1186/s40959-024-00295-1
  11. 11. Chen M-Y, Wen J-X, Lu M-T, Jian X-Y, Wan X-L, Xu Z-W. Association between prognostic nutritional index and prognosis in patients with heart failure: a meta-analysis. Front Cardiovasc Med. (2022) 9: 918566. doi:, doi:10.3389/fcvm.2022.918566
  12. 12. Banerjee A, Dashtban A, Chen S, Pasea L, Thygesen JH, Fatemifar G. Identifying subtypes of heart failure from three electronic health record sources with machine learning: an external, prognostic, and genetic validation study. Lancet Digit Health. (2023) 5: e370 – 9. doi:, doi:10.1016/S2589-7500(23)00065-1
  13. 13. Sheng Y, Liao X, Li F, Xu S, Zhang D. Clinical subtypes and prognosis of new-onset atrial fibrillation in critically ill patients. Int J Cardiol Cardiovasc Risk Prev. (2026) 30: 200654. doi:, doi:10.1016/j.ijcrp.2026.200654
  14. 14. Garcia Acevedo G, Ahmad A, Stall B, Mokhtarnia M, Lapp JM, Verma AA. International comparison of quality indicators for adults hospitalized for heart failure: a systematic review. Circ Cardiovasc Qual Outcomes. (2024) 17: e010629. doi:, doi:10.1161/CIRCOUTCOMES.123.010629
  15. 15. Fryar J, Khawaja S, Dang T, Chan W, Bayat MK, Parsonage W. Timing, diagnosis, and potential preventability of 30-day unplanned readmissions after a heart failure hospitalisation: implications for care quality. Heart Lung Circ. (2025) 34: 1281 – 9. doi:, doi:10.1016/j.hlc.2025.04.085
  16. 16. Gevaert AB, Tibebu S, Mamas MA, Ravindra NG, Lee SF, Ahmad T. Clinical phenogroups are more effective than left ventricular ejection fraction categories in stratifying heart failure outcomes. ESC Heart Fail. (2021) 8: 2741 – 54. doi:, doi:10.1002/ehf2.13344
  17. 17. Murray EM, Greene SJ, Rao VN, Sun J-L, Alhanti BA, Blumer V. Machine learning to define phenotypes and outcomes of patients hospitalized for heart failure with preserved ejection fraction: findings from ASCEND-HF. Am Heart J. (2022) 254: 112 – 21. doi:, doi:10.1016/j.ahj.2022.08.009
  18. 18. Choudhary T, Upadhyaya P, Davis CM, Yang P, Tallowin S, Lisboa FA. Derivation and validation of generalized sepsis-induced acute respiratory failure phenotypes among critically ill patients: a retrospective study. Crit Care. (2024) 28: 321. doi:, doi:10.1186/s13054-024-05061-4
  19. 19. Gorący I, Rębacz-Maron E, Korbecki J, Gorący J. Concentrations of Mg, ca, Fe, cu, Zn, P and anthropometric and biochemical parameters in adults with chronic heart failure. PeerJ. (2021) 9: e12207. doi:, doi:10.7717/peerj.12207
  20. 20. Nouraei H, Nouraei H, Rabkin SW. Comparison of unsupervised machine learning approaches for cluster analysis to define subgroups of heart failure with preserved ejection fraction with different outcomes. Bioengineering (Basel). (2022) 9: 175. doi:, doi:10.3390/bioengineering9040175
  21. 21. Sung K-T, Chang H-Y, Hsu N-W, Huang W-H, Lin Y-H, Yun C-H. Penalized model-based unsupervised phenomapping unravels distinctive HFrEF phenotypes with improved outcomes discrimination from sacubitril/valsartan treatment independent of MAGGIC score. J Am Heart Assoc. (2023) 12: e028860. doi:, doi:10.1161/JAHA.122.028860
  22. 22. Pan J, Liu M, Huang J, Chen L, Xu Y. Impact of anemia on clinical outcomes in patients with acute heart failure: a systematic review and meta-analysis. Clin Cardiol. (2024) 47: e24228. doi:, doi:10.1002/clc.24228
  23. 23. Xanthopoulos A, Giamouzis G, Dimos A, Skoularigki E, Starling RC, Skoularigis J. Red blood cell distribution width in heart failure: pathophysiology, prognostic role, controversies and dilemmas. J Clin Med. (2022) 11: 1951. doi:, doi:10.3390/jcm11071951
  24. 24. Beavers CJ, Ambrosy AP, Butler J, Davidson BT, Gale SE, Piña IL. Iron deficiency in heart failure: a scientific statement from the heart failure society of America. J Card Fail. (2023) 29: 1059 – 77. doi:, doi:10.1016/j.cardfail.2023.03.025
  25. 25. Liu Z, Nguyen Khuong J, Borg Caruana C, Jackson SM, Campbell R, Ramson DM. The prognostic value of elevated perioperative neutrophil-lymphocyte ratio in predicting postoperative atrial fibrillation after cardiac surgery: a systematic review and meta-analysis. Heart Lung Circ. (2020) 29: 1015 – 24. doi:, doi:10.1016/j.hlc.2019.11.021
  26. 26. Zhang C, Li M, Liu L, Deng L, Yulei X, Zhong Y. Systemic immune-inflammation index as a novel predictor of major adverse cardiovascular events in patients undergoing percutaneous coronary intervention: a meta-analysis of cohort studies. BMC Cardiovasc Disord. (2024) 24: 189. doi:, doi:10.1186/s12872-024-03849-4
  27. 27. Young JB, Eknoyan G. Cardiorenal syndrome: an evolutionary appraisal. Circ Heart Fail. (2024) 17: e011510. doi:, doi:10.1161/CIRCHEARTFAILURE.123.011510
  28. 28. Ghazal M, Khalife WI. Hypoalbuminemia in heart failure: pathophysiology, clinical implications, and management strategies. Heart Fail Rev. (2025) 30: 1407 – 14. doi:, doi:10.1007/s10741-025-10558-3
  29. 29. Qin S, Xiang M, Gao L, Cheng X, Zhang D. Uric acid is a biomarker for heart failure, but not therapeutic target: result from a comprehensive meta-analysis. ESC Heart Fail. (2024) 11: 78 – 90. doi:, doi:10.1002/ehf2.14535
  30. 30. Kurniawan RB, Oktafia P, Saputra PBT, Purwati DD, Saputra ME, Maghfirah I. The roles of C-reactive protein-albumin ratio as a novel prognostic biomarker in heart failure patients: a systematic review. Curr Probl Cardiol. (2024) 49: 102475. doi:, doi:10.1016/j.cpcardiol.2024.102475
  31. 31. Chen W, Zhang G, Lei Q, Lu H. Prognostic value of advanced lung cancer inflammation index in heart failure patients: a comprehensive analysis. ESC Heart Fail. (2025) 12: 2298 – 309. doi:, doi:10.1002/ehf2.15178
  32. 32. Shi T, Wang Y, Peng Y, Wang M, Zhou Y, Gu W. Advanced lung cancer inflammation index combined with geriatric nutritional risk index predict all-cause mortality in heart failure patients. BMC Cardiovasc Disord. (2023) 23: 565. doi:, doi:10.1186/s12872-023-03608-x
  33. 33. Foroutan F, Rayner DG, Ross HJ, Ehler T, Srivastava A, Shin S. Global comparison of readmission rates for patients with heart failure. J Am Coll Cardiol. (2023) 82: 430 – 44. doi:, doi:10.1016/j.jacc.2023.05.040
  34. 34. Jain A, Arora S, Patel V, Raval M, Modi K, Arora N. Etiologies and predictors of 30-day readmission in heart failure: an updated analysis. Int J Heart Fail. (2023) 5: 159 – 68. doi:, doi:10.36628/ijhf.2023.0015
  35. 35. Talha KM, Pandey A, Fudim M, Butler J, Anker SD, Khan MS. Frailty and heart failure: state-of-the-art review. J Cachexia Sarcopenia Muscle. (2023) 14: 1959 – 72. doi:, doi:10.1002/jcsm.13306
  36. 36. Davis E, Dunbar S, Higgins M, Wood K, Ferranti E, Morris A. Heart failure symptom burden, dietary intake, and inflammation: an integrative review of the literature. J Integr Nurs. (2023) 5: 81 – 92. doi:, doi:10.4103/jin.jin_26_23
  37. 37. Chaikijurajai T, Tang WHW. Gut microbiome and precision nutrition in heart failure: hype or hope? Curr Heart Fail Rep. (2021) 18: 23 – 32. doi:, doi:10.1007/s11897-021-00503-4
  38. 38. Bilicki DJ, Reeves MJ. Outpatient follow-up visits to reduce 30-day all-cause readmissions for heart failure, COPD, myocardial infarction, and stroke: a systematic review and meta-analysis. Prev Chronic Dis. (2024) 21: E74. doi:, doi:10.5888/pcd21.240138
  39. 39. Tyler N, Hodkinson A, Planner C, Angelakis I, Keyworth C, Hall A. Transitional care interventions from hospital to community to reduce health care use and improve patient outcomes: a systematic review and network meta-analysis. JAMA Netw Open. (2023) 6: e2344825. doi:, doi:10.1001/jamanetworkopen.2023.44825

Citation

Zhang, Y., et al. (2026). Laboratory-derived inflammatory–nutritional exploratory phenogroups and their associations with functional limitation and 30-day same-center all-cause unplanned readmission in patients with heart failure. Frontiers in Nutrition, Volume 13. https://doi.org/
Received
Accepted
Published
Volume
13 - 2026

Copyright

Copyright © 2026 Zhang, Zhang, Song, Hu and Nan.

License: https://creativecommons.org/licenses/by/4.0/

*Correspondence: Jingzhu Nan

Article metrics

Views
0
Downloads
0
Citations
0
Shares
0

Related articles

Mini ReviewIron homeostasis at the interface of nutrition, infection, and iron supplementationÖzge Mert, Seray Saday, Aybala TemelClinical TrialQuality of life and safety of a dietary intervention with French maritime pine bark extract (Pycnogenol ®) in women with osteoporotic bone loss: a randomized, placebo-controlled double-blind study—Part IIChristian Scherg, Franca Genest, Jonas Maghary, Lothar Seefried, Petra HöggerOriginal ResearchBody weight–normalized resting metabolic rate and its associations with malnutrition and sarcopenia in hospitalized older adults: a cross-sectional studyShan Jiang, Jie Zhao, Peng Chen, Lehui Chen, Weihong Chen, Zi Lin, Lin Kang, Hailong Li, Xia Zhao, Qiumei Wang