What happened
The November 2026 issue of the Journal of Cardiac Failure, the official journal of the Heart Failure Society of America, published a study showing that adding an artificial‑intelligence analysis of a single routine echocardiogram clip (the EchoGo HF model) to the H₂FPEF clinical score identified 78% of patients with hemodynamically confirmed heart‑failure‑preserved ejection fraction (HFpEF). This compares with a 48% detection rate using the H₂FPEF score alone. The study evaluated 280 outpatients referred for suspected HFpEF, each of whom also underwent exercise hemodynamic testing, the diagnostic gold standard. The AI model added the most diagnostic value for patients with low or intermediate H₂FPEF scores, prompting a two‑step strategy: calculate the H₂FPEF score first, then apply the AI model only when the score is inconclusive. An AI‑detected HFpEF result also independently predicted elevated filling pressures and reduced exercise capacity, suggesting the model reflects disease severity as well as diagnosis. The issue also featured a large‑language‑model analysis (using OpenAI’s o1) of why many heart‑failure patients are not prescribed guideline‑recommended SGLT2 inhibitors, achieving 84.7% accuracy in classifying documented reasons, with physician‑attributed factors most common. A review article mapped AI applications across the heart‑failure care continuum, concluding that AI integrated into active clinical workflows shows the strongest evidence of benefit.
The Journal of Cardiac Failure’s November 2026 focus issue on artificial intelligence presented a study where researchers applied the EchoGo HF AI model to resting echocardiograms from 280 outpatients suspected of HFpEF. Each patient also underwent exercise hemodynamic testing, the gold‑standard diagnostic method. When the AI analysis was combined with the H₂FPEF clinical score, the detection rate for hemodynamically confirmed HFpEF rose to 78%, compared with 48% using the score alone.
The AI model proved most valuable for patients with low or intermediate H₂FPEF scores—cases where diagnosis is traditionally hardest. The authors therefore proposed a two‑step diagnostic pathway: first calculate the H₂FPEF score, then apply the AI model only when the score is inconclusive. This approach aims to limit AI use to cases where it adds the greatest incremental benefit.
Beyond diagnosis, the AI‑detected HFpEF result independently predicted elevated filling pressures and reduced exercise capacity, indicating that the model captures disease severity. The study authors suggest that such prognostic information could help tailor treatment intensity.
The issue also included a separate investigation using OpenAI’s o1 large language model to analyze clinical notes from 599 cardiology patients. The model classified documented reasons for not prescribing SGLT2 inhibitors with 84.7% accuracy. Physician‑attributed reasons accounted for 28.7% of non‑prescriptions, with the most common specific reasons being deferral to a future visit (15.4%) and physician judgment that the patient was low‑risk or unlikely to benefit (7.4%).
A review article in the same issue mapped AI applications across the heart‑failure care continuum, concluding that AI tools embedded in active clinical workflows—such as guiding targeted testing, treatment optimization, and follow‑up—have the strongest evidence of benefit, whereas passive alert systems show weaker impact.
Source details: prnewswire.com
Why it matters
Improving HFpEF diagnosis is clinically significant because HFpEF accounts for roughly half of all heart‑failure cases, yet it remains under‑diagnosed due to the need for invasive hemodynamic testing. By raising detection from 48% to 78% with a non‑invasive, AI‑enhanced echocardiogram, the EchoGo HF model could reduce reliance on costly, invasive procedures, accelerate appropriate treatment, and potentially improve outcomes for a large patient population. The study also highlights how AI can augment existing risk scores, targeting the diagnostic gray zone where clinicians struggle most. Additionally, the large‑language‑model analysis of SGLT2‑inhibitor prescribing gaps identifies concrete, physician‑driven barriers, offering actionable insights to improve guideline adherence and patient care. Together, these findings illustrate a shift from retrospective, in‑silico AI demonstrations toward tools that may be integrated into real‑world clinical workflows, addressing both diagnostic and therapeutic challenges in heart failure.
HFpEF is a growing public‑health challenge, representing about 50% of heart‑failure cases and associated with high morbidity, mortality, and health‑care costs. Accurate, non‑invasive diagnosis is essential for timely therapy, yet current diagnostic pathways often require invasive hemodynamic testing, limiting accessibility.
The EchoGo HF model’s ability to raise detection rates to 78% using a single routine echocardiogram clip could dramatically reduce the need for invasive testing, lower costs, and speed up treatment initiation. This is especially relevant given the shortage of heart‑failure specialists and unfilled training positions noted by the journal’s editors.
The large‑language‑model analysis of SGLT2‑inhibitor prescribing gaps provides concrete, data‑driven targets for quality‑improvement initiatives. Since SGLT2 inhibitors improve outcomes in heart‑failure patients, addressing physician‑driven barriers could increase guideline‑concordant care and improve survival.
Both studies exemplify a shift from purely retrospective AI performance reports toward research that directly addresses clinical decision points, disease severity assessment, and real‑world prescribing behavior, aligning with calls from the Heart Failure Society of America for prospective, outcome‑focused AI validation.
If validated prospectively, these AI tools could be incorporated into electronic health‑record systems, offering clinicians decision support at the point of care, potentially improving diagnostic accuracy, treatment adherence, and overall patient outcomes.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Future prospective trials that test the EchoGo HF model in real‑time clinical settings, including its impact on patient outcomes, cost‑effectiveness, and workflow integration. Monitoring regulatory review or FDA clearance pathways for AI‑based diagnostic software will indicate how quickly the technology could reach broader use. Follow‑up studies on the large‑language‑model approach to prescribing behavior may lead to decision‑support tools that clinicians to consider SGLT2 inhibitors when appropriate. Finally, watch for adoption guidelines from professional societies such as the HFSA that could standardize AI use in heart‑failure care.
Prospective, multicenter trials evaluating the EchoGo HF model’s impact on diagnostic accuracy, patient outcomes, and health‑care utilization.
Regulatory pathways for AI‑based diagnostic software, including FDA clearance or CE marking, which will affect market entry timelines.
Development of clinical decision‑support interfaces that integrate the EchoGo HF model into existing echocardiography workflows without adding significant clinician burden.
Follow‑up research on the large‑language‑model approach to prescribing, including pilot implementations of real‑time alerts or recommendation engines within cardiology EHRs.
Guidance documents or consensus statements from the Heart Failure Society of America or other professional bodies that define best practices for AI integration in heart‑failure diagnosis and management.