AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
Impact of AI-Enabled Detection of HFpEF on Treatment Initiation, Clinical, and Economic Outcomes
By Akerman A, MacDonald A, Dermon J et al.
Background: Heart failure with preserved ejection fraction (HFpEF) is common and frequently unrecognized. Earlier recognition may enable more timely treatment and reduce healthcare utilization, yet evidence linking AI-enabled detection to downstream outcomes and economic impact is limited. Hypothesis: Utilizing an AI platform for HFpEF detection would demonstrate downstream clinical and economic impact when evaluated using the AHA AI Assessment Lab, powered by Dandelion Health platform. Methods: A retrospective cohort study of adults from hospitals across 2 US health systems examined patients undergoing transthoracic echocardiography (TTE) between 2018-2024 with preserved EF (>40%), no prior heart failure (HF) diagnosis (ICD-10, I50.x), but who developed a new diagnosis of HF following the TTE. Patients who were identified at the point of the TTE (or within 90 days) by “standard-of-care” had 3 or more of the following: newly initiated HF therapy, relevant HF ICD code, elevated (NT)pro-BNP, or elevated filling pressure on TTE. Patients who were not identified by standard-of-care (less than 3 of these criteria) but were positive by EchoGo Heart Failure (Ultromics Ltd) and were probable risk of HFpEF (H2FPEF Score), are represented as “AI-enhanced workflow”. Outcomes compared between the groups included time to HFpEF-relevant GDMT initiation, MACE-related hospitalisations, 30-day readmissions, and emergency department (ED) visits. A severity-adjusted Markov model estimated health system economic impact using observed outcomes and publicly available cost inputs. Results: Standard of care identified 1,152 patients who developed HF after their index TTE (69%; 74 ± 12 yrs, 51% female, 11% identified non-white race) and the AI-enhanced workflow captured an additional 515 (31%; 76 ± 8 yrs, 50% female, 7% identified non-white race). Patients who would have been identified by the AI-enhanced workflow but were missed by standard-of-care, had HFpEF-relevant therapies initiated later (median difference: MRA 247 days; ARNI 276 days; SGLT2i 283 days), higher mortality (1.1% vs 0.5% per month; annual absolute difference 6.7%), and greater acute care utilization. In economic modelling, the AI-enhanced workflow would have been associated with 245 fewer MACE-related 30-day readmissions, 280 fewer MACE-related ED visits and 237 fewer deaths, with an average cost saving of $363 per patient versus routine care per health system annually. Conclusion: Patients who would have been identified by the AI-enhanced workflow but missed by standard-of-care within 90 days of their index TTE incur excess resource utilization and poorer outcomes. Implementing an AI-enhanced workflow at the time of the index TTE could therefore lead to earlier uptake of HFpEF-relevant therapies, improved downstream outcomes, and favourable health system economic impact.