Study identifies two distinct HFpEF subtypes with different disease mechanisms

Heart failure with preserved ejection fraction—a condition where the heart pumps normally on paper but leaves patients breathless and exhausted—has long resisted simple explanation. Millions of people worldwide carry this diagnosis, yet doctors struggle to treat it effectively because the disease wears different masks in different bodies. A new study of 84 patients offers a way to see through the confusion: there are not one but two distinct versions of this illness, each driven by different mechanisms, each demanding different approaches.

Researchers at Qingdao University recruited patients between December 2023 and December 2024, collecting detailed heart imaging and blood work. Using unsupervised machine learning—a technique that finds natural groupings without being told what to look for—they identified two clusters that split the patient population almost evenly. The first group, 43 patients, were younger (average age 58.6 years) and carried a heavy burden of kidney disease. Their kidneys filtered blood at a rate of just 12.8 milliliters per minute per 1.73 square meters of body surface—severe dysfunction by any measure. Their hearts showed pronounced mechanical dyssynchrony, meaning different regions of the heart muscle contracted out of sync, like an orchestra where the violins lag behind the drums. The second group, 41 patients, were older (average 71.2 years) with kidneys that functioned normally. Their hearts showed less mechanical discord and different patterns of stress hormones in the blood.

The study's central finding concerns a measurement called peak strain dispersion—essentially a way of quantifying how well different parts of the heart muscle work together. The researchers used advanced machine learning models to understand how this measurement acts as a bridge between upstream problems (kidney function, hormone levels) and downstream cardiac dysfunction (wasted energy, reduced efficiency). What emerged was striking: peak strain dispersion mattered profoundly, but in opposite ways for the two groups. In younger patients with kidney disease, kidney function showed a trend toward influencing heart efficiency through mechanical dyssynchrony, though the effect fell short of statistical significance. In older patients, by contrast, elevated B-type natriuretic peptide—a hormone released when the heart is stressed—had its effects on heart function almost entirely channeled through mechanical dyssynchrony. The younger group's disease seemed to involve multiple pathways beyond simple mechanical discord; the older group's disease appeared to flow directly through it.

The researchers employed a technique called SHAP (SHapley Additive exPlanations) to peer inside their machine learning models and understand which features mattered most. Peak strain dispersion emerged as the single most important predictor of myocardial work parameters. When they added interaction terms—combinations of variables that capture how one factor's effect depends on another—the picture grew richer. The interaction between peak strain dispersion and left ventricular ejection fraction ranked among the top predictors of heart efficiency, suggesting that mechanical dyssynchrony hits hardest in hearts that are already weakened.

These findings carry immediate clinical weight. They suggest that HFpEF is not a single disease requiring a universal treatment but rather a family of related conditions, each with its own dominant mechanism. For younger patients trapped in cardiorenal syndrome—where failing kidneys and failing hearts feed each other's decline—treatment should prioritize kidney protection and careful fluid management. For older patients whose disease is driven primarily by age-related changes and neurohormonal activation, therapies aimed at improving heart muscle synchronization might prove more effective. The study's machine learning models predicted myocardial work parameters with moderate accuracy (explaining 58 to 61 percent of the variance), a performance the authors acknowledge reflects the genuine complexity of the disease—unmeasured factors like genetic predisposition, the microbiome, and dynamic changes over time all play roles that a single snapshot cannot capture.

The work rests on solid ground. The researchers validated their measurements of peak strain dispersion in a subset of patients, finding excellent reproducibility (intraclass correlation coefficients of 0.98 and 0.97 in the two clusters). They acknowledge limitations: this is a single-center study with a relatively small sample, the population was predominantly Chinese, and the cross-sectional design cannot establish causality. Yet the findings align with recent research on HFpEF heterogeneity and offer a concrete framework for moving toward personalized medicine. If these results hold in larger, more diverse populations, they could reshape how doctors approach a disease that has resisted simple solutions for decades.