The metabolic-inflammatory axis in chronic heart failure: integrating the NLR and TyG index for recent-onset atrial fibrillation prediction via machine learning and mediation analysis.
Yao C, Liu X, et al. • Frontiers in endocrinology • 2026
An interpretable machine learning framework identified the immune-metabolic axis (NLR and TyG index) as an important predictive component of recent-onset atrial fibrillation in chronic heart failure, with the Extra Trees model achieving an AUC of 0.853 in training and 0.766 in external validation.
Key Findings
Results
The Extra Trees machine learning model performed best among nine algorithms for predicting recent-onset atrial fibrillation in chronic heart failure patients.
The Extra Trees model achieved an area under the receiver operating characteristic curve (AUC) of 0.853 in the training cohort.
External validation in 277 additional patients yielded an AUC of 0.766.
Nine ML algorithms were compared in total.
Model interpretability was assessed using SHapley Additive exPlanations (SHAP).
The study included 4,872 hospitalized patients with CHF from Guang'anmen Hospital and Xiyuan Hospital in the primary cohort.
Results
The neutrophil-to-lymphocyte ratio (NLR) demonstrated a nonlinear association with atrial fibrillation risk, with a steeper increase in risk below a threshold of 4.24.
Segmented regression was used to examine nonlinear threshold effects.
The NLR threshold was identified at 4.24, below which the association with AF risk was steeper.
NLR was identified as one of the key predictive variables in the machine learning model.
NLR is described as an inexpensive, clinically accessible biomarker.
Results
The triglyceride-glucose (TyG) index showed a threshold-dependent J-shaped relationship with atrial fibrillation risk, with risk increasing significantly above 5.91.
The TyG index threshold was identified at 5.91, above which AF risk increased significantly.
The relationship was characterized as J-shaped.
TyG index was analyzed alongside demographic, clinical, and laboratory variables.
The combination of TyG and NLR further improved model discrimination beyond either marker alone.
Results
Mediation analysis suggested that NLR mediates a substantial proportion of the observed association between the TyG index and atrial fibrillation.
The estimated indirect association through NLR accounted for a substantial proportion of the observed association between TyG and AF.
This finding supports an immunometabolic pathway linking metabolic dysregulation (TyG) through inflammatory dysregulation (NLR) to AF risk.
Mediation analysis was performed specifically to explore the immunometabolic pathway linking TyG, NLR, and AF.
The finding identifies a high-risk phenotype characterized by concurrent metabolic stress and inflammation.
Methods
The study was designed as a retrospective multicenter study including 4,872 hospitalized CHF patients with external validation in 277 additional patients.
Primary cohort patients were drawn from Guang'anmen Hospital and Xiyuan Hospital.
External validation was performed in 277 additional patients.
Demographic, clinical, and laboratory variables were analyzed.
The study was registered in the Chinese Trial Registry (ChiCTR, ITMCTR2025001576).
The design was retrospective and multicenter.
Conclusions
NLR and TyG index were identified as inexpensive, clinically accessible biomarkers that may improve early risk stratification for recent-onset AF in CHF.
Both markers are described as inexpensive and clinically accessible.
They may identify a high-risk phenotype characterized by concurrent metabolic stress and inflammation.
The combination of both markers improved model discrimination compared to individual markers.
The study positioned these biomarkers as potentially useful for early risk stratification in clinical practice.
What This Means
This research suggests that two simple, inexpensive blood tests — the triglyceride-glucose (TyG) index, which reflects metabolic health and insulin resistance, and the neutrophil-to-lymphocyte ratio (NLR), which reflects inflammation — can together help predict which patients hospitalized with chronic heart failure are likely to develop a new episode of atrial fibrillation (an irregular heart rhythm). The researchers used advanced computer algorithms (machine learning) trained on nearly 5,000 patients across two hospitals, and tested the best model on an additional 277 patients. The best-performing algorithm, called Extra Trees, correctly identified at-risk patients with reasonable accuracy in both the original and the separate validation group.
The study also found that the relationships between these biomarkers and atrial fibrillation risk are not simple straight lines. NLR showed a steeper rise in AF risk at lower values (below 4.24), while TyG showed a J-shaped curve where risk jumped above a value of 5.91. Importantly, a statistical technique called mediation analysis suggested that inflammation (measured by NLR) may act as a connecting pathway between metabolic stress (measured by TyG) and atrial fibrillation — meaning metabolic problems may partly trigger AF by first driving up inflammation.
This research suggests that checking these two routine lab values together could help clinicians identify heart failure patients who are at higher risk of developing atrial fibrillation before it occurs, potentially enabling earlier monitoring or intervention. Because both tests are already routinely available in hospitals, implementing this type of risk stratification would not require new or expensive testing infrastructure.
Yao C, Liu X, Liu R, Su D, Yang K, Yao L, et al.. (2026). The metabolic-inflammatory axis in chronic heart failure: integrating the NLR and TyG index for recent-onset atrial fibrillation prediction via machine learning and mediation analysis.. Frontiers in endocrinology. https://doi.org/10.3389/fendo.2026.1831904