Peak strain dispersion as a nonlinear mediator in HFpEF: Unraveling subtype-specific pathways via SHAP-augmented ensemble modeling

Mingming Lin, Kai Li, Xiaofan Wang, Juanjuan Sun, Kun Gong, Zhibin Wang and Pin Sun

PLOS Computational Biology, 2026, vol. 22, issue 1, 1-15

Abstract: Background: Heart failure with preserved ejection fraction (HFpEF) represents a heterogeneous syndrome with diverse pathophysiological mechanisms and limited therapeutic options. Peak strain dispersion (PSD) has emerged as a potential mediator in HFpEF pathophysiology. This study aimed to identify distinct HFpEF subtypes and investigate PSD’s subtype-specific mediating pathways. Methods: This prospective single-center study included 150 HFpEF patients recruited from December 2023 to December 2024. Unsupervised K-means clustering was performed on the entire cohort to identify patient subtypes. For detailed analysis, rigorous data quality control was performed by removing cases with missing values in any of the 25 baseline features or outcome variables. Consequently, 84 patients with complete data were retained for analysis. Comprehensive clinical and echocardiographic data were collected, including PSD measured by speckle-tracking echocardiography and myocardial work parameters (global work waste and global work efficiency). Unsupervised K-means clustering was performed to identify distinct patient subtypes using eight key variables. Machine learning models with feature engineering (incorporating five clinically meaningful interaction terms: PSD_LVEF, age_HTN, eGFR_BNP, RWT_E/e’, and GLS_LVMI) were developed to predict myocardial work parameters and assess feature importance using SHAP (SHapley Additive exPlanations) analysis. Nonlinear mediation analysis was conducted within each subtype to evaluate the mediating pathways through which clinical factors influence myocardial work outcomes. Results: Two distinct HFpEF subtypes were identified: Cluster 0 characterized by younger age (58.6 ± 13.2 years), severe renal dysfunction (eGFR 12.8[8.9-19.9] mL/min/1.73m²), higher PSD (56.0[48.0-64.5] ms), and lower global work efficiency; and Cluster 1 characterized by older age (71.2 ± 9.7 years), preserved renal function (eGFR 104.0[78.5-126.0] mL/min/1.73m²), lower PSD (41.0[35.0-49.0] ms), and higher GWE. Machine learning models achieved moderate to good predictive performance (R² = 0.58-0.61 for GWE and GWW). SHAP analysis revealed that PSD was the most important predictor, with the PSD×LVEF interaction term showing prominent importance in GWE prediction. Nonlinear mediation analysis demonstrated striking subtype-specific differences in mediation patterns.In Cluster 0, eGFR showed a trend toward mediating its effects on GWW through PSD (indirect effect = 0.313), reflecting complex cardiorenal interactions in younger patients with severe renal disease. In contrast, Cluster 1 demonstrated significant mediation effects: BNP’s effect on GWW was significantly mediated through PSD (indirect effect = -0.4877, P

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1013891 (text/html)
https://journals.plos.org/ploscompbiol/article/fil ... 13891&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pcbi00:1013891

DOI: 10.1371/journal.pcbi.1013891

Access Statistics for this article

More articles in PLOS Computational Biology from Public Library of Science
Bibliographic data for series maintained by ploscompbiol ().