Autonomic neuromodulation guided by machine learning for phenotype-specific HFpEF management

To The Editor, Heart failure with preserved ejection fraction (HFpEF) is a syndrome where the heart cannot fill properly even though the left ventricular ejection fraction (EF) is ≥50%. HFpEF accounts for about half of all heart failure cases worldwide and affects people with hypertension, obesity, and diabetes. Patients often have symptoms of dyspnea (shortness of breath) and exercise intolerance. HFpEF arises from stiffening of the heart muscle, abnormal relaxation, endothelial dysfunction, and systemic inflammation. It results in elevated filling pressures and repeated hospital admissions. Patients hospitalized with HFpEF have an annual mortality rate of around 15%. Diagnosis typically requires echocardiography showing preserved EF plus evidence of elevated filling pressures or diastolic dysfunction. Treatment options remain limited despite guideline-directed medical therapies, and mortality remains high. Comorbid conditions such as chronic kidney disease worsen outcomes. HFpEF represents a significant unmet need in heart failure care and calls for novel therapeutic strategies1,2 Neuromodulation is an interventional technique that aims to rebalance the autonomic nervous system in HFpEF. Techniques include vagus nerve stimulation (VNS), baroreceptor activation therapy (BAT), and greater splanchnic nerve (GSN) ablation. These approaches target excessive sympathetic activity and reduced parasympathetic tone. Artificial intelligence (AI) enhances this process by identifying specific phenotypes within HFpEF patients who may respond best to particular neuromodulation strategies. AI models use clinical, imaging, and physiological data to stratify patients and optimize stimulation parameters. The working principle of neuromodulation is to restore autonomic balance, improve cardiac function, and reduce inflammation. AI-guided neuromodulation is also being explored in related conditions such as hypertension and HFrEF3. A sham-controlled randomized trial showed that low-level transcutaneous VNS improved global longitudinal strain and reduced tumor necrosis factor-α levels after 3 months in HFpEF patients, with no device-related side effects. Adherence to daily stimulation was over 90%, and quality of life improved significantly. Animal studies also support neuromodulation’s benefits, showing decreased inflammation, less fibrosis, and improved cardiac function in HFpEF models. Another review of neuromodulation therapies (VNS, BAT, and GSN ablation) in HFpEF highlighted potential improvements in symptoms, exercise capacity, and quality of life, though individual results varied. These findings suggest that neuromodulation can beneficially alter autonomic tone and clinical outcomes in HFpEF, providing evidence for further research and personalized interventions1,4 Despite the promise, challenges remain. First, neuromodulation carries procedural risks and device-related adverse events. Safety profiles vary between methods, and long-term data are limited. Second, AI models need large, diverse datasets for training and validation to avoid biased predictions and ensure reproducibility. Third, clinician awareness of neuromodulation options in HFpEF is currently low, slowing adoption and limiting clinical implementation. Cost and limited access to advanced neuromodulatory devices also hinder widespread use. Furthermore, standardized protocols for AI-guided therapy are lacking, and regulatory frameworks for AI integration in clinical care remain undeveloped5. In conclusion, AI-guided autonomic neuromodulation represents a novel frontier for HFpEF management. It combines advanced analytics with targeted therapy to tailor interventions for individuals. Future work should focus on large randomized trials, real-world validation of AI models, clinician education, and cost reduction strategies. Increased awareness and collaborative research will help move this promising approach into wider clinical practice. This letter to the editor adheres to the Transparency in the Reporting of AI in Research (TITAN) guideline6. Sincerely,